Block Reference¶

Auto-generated parameter reference for all pipeline blocks. Highlighted parameters are primary — they directly affect output count, quality, or runtime. Computed defaults resolve from settings or model logic at runtime; no documentation builder's site configuration is embedded here.

Resources list what a block requests: partition is the class cpu or gpu, and gpu_class is the GPU size (small or large). Each site maps these to a real partition and --gres through its CHAPERONE_* settings; see Slurm Site Configuration.

Network Blocks

Choose a predictor whose supported molecules, inputs, and validation evidence fit the task. Listing a block here does not establish scientific equivalence or a successful fresh-cluster installation. See Core Examples for benchmark scope and Validation for prediction comparisons.

esmfold2 — ESMFold2 structure prediction with optional MSA conditioning
openfold3 — OpenFold 3 structure prediction
boltz2 — Boltz-2 structure prediction [paper]
boltz2_steered — Boltz-2 with helix steering (Protein Hunter)
rf3 — RosettaFold3 structure prediction [paper]
protenix — Protenix structure prediction
chai1 — Chai-1 structure prediction
af3 — AlphaFold3 structure prediction [paper]
af3_with_target_template — AlphaFold3 with the target chain structure as a fixed template
colabfold — ColabFold (AlphaFold2) structure prediction
bindcraft — AlphaFold2-guided de novo protein binder design [paper]
bindcraft2 — JAX binder optimization, ProteinMPNN redesign, and AlphaFold2 validation
boltzgen — BoltzGen protein design block [paper]
disco — Multimodal protein sequence-structure co-design
e1 — E1 masked language model for sequence prediction [paper]
esm2 — ESM-2 protein language model for sequence embeddings [paper]
foundry_mpnn — Foundry MPNN for inverse folding [paper]
manifest_val_structures — Manifest reference PDB assemblies for prediction validation
mpnn — Inverse folding with LigandMPNN (protein, ligand, soluble and membrane models) [paper]
protein_hunter — De novo binder design via Boltz structure hallucination [paper]
proteina_complexa — All-atom protein binder sequence/structure co-design [paper]
rfd1 — RFdiffusion v1 backbone-level protein design
rfd2 — RFdiffusion2 (rf_diffusion) block for atomic-level protein design
rfd3 — RFdiffusion3 block [paper]
rfd3_partial — RFdiffusion3 partial diffusion
rfd3na — RFdiffusion3NA block (all-atom protein/DNA/RNA design) [paper]
rfoptimization — RF3 gradient / Boltz-2 cycling with external MPNN sequence redesign [paper]
rpxdock — Dock a z-oriented C3 asymmetric unit into an I3 cage and emit C3 trimers
score_val_structures — Score predictions against manifested validation references
simplefold — Apple SimpleFold sequence-only protein monomer prediction
tmol_interface — tmol protein-ligand interface ddG + per-term breakdown
tmol_minimize — tmol Cartesian all-atom energy minimisation (beta_nov2016_cart)

Metric Blocks

aromatic_neighbors — Count aromatic residues near ligand atoms
chirality_check — Per-atom CIP-code chirality comparison vs a reference ligand
clash — Count non-bonded backbone heavy-atom clashes
contact_ms — Compute contact molecular surface between binder and target chains
dssr_motif_base_pairing — Mapped motif base pairing and native DSSR classification retention.
dssr_snap_metrics — Per-design biophysical metrics for protein-DNA complexes via x3dna-dssr and x3dna-snap (3DNA v2.3).
foldseek — Cluster structures by structural similarity using Foldseek [paper]
hbond_metrics — Compute hydrogen bond metrics for ligand interactions
ipsae — Interface predicted Score from Aligned Errors
ligand_clash — Measure the closest protein-ligand heavy-atom contact
ligand_ddg — Compute Rosetta DDG binding energy
ligand_ddg_soft — Soft-repulsive ddG (pack hard, score soft, trimmed mean)
ligand_pocket — Distributed ligand contact, enclosure, and shell metrics
metal3d — Predict transition-metal binding sites with the Metal3D 3D-CNN
motif_confidence — Motif/scaffold pLDDT and PAE split, via diffused_index_map
motif_geometry — Distances/angles between motif atoms, via diffused_index_map
mpnn_score — Parse MPNN FASTA confidences as ranking metrics
peppr — Compute structural evaluation metrics using peppr
plip — Profile protein-ligand interactions using PLIP
pocket_coulomb — Electrostatic complementarity between charged ligand atoms and charged pocket residues
ppi_ddg — Protein-protein binding energy with FastRelax + ddG
rasa — Compute solvent accessible surface area for ligand atoms
reference_motif_rmsd — Motif RMSD against the input reference, via diffused_index_map
rmsd — Compute alignment RMSD metrics between two designs
rna_connectivity — O3'(i)->P(i+1) backbone breaks in nucleic-acid chains
rna_ss_f1_mcc — Motif-restricted secondary-structure F1 and MCC of an RNet2 prediction against the reference's canonical pairs
rna_thermo — ViennaRNA Boltzmann probability, pair probabilities and ensemble defect for a required motif pair set
rotamer_metrics — Compute rotamer quality metrics for protein sidechains
secondary_structure — Per-fold SS composition + loop/pLDDT disorder, globally and pocket-around-ligand
sequence_identity_gate — Global NW identity of the scaffold region against natural references; hard novelty gate
symmetry — Cyclic point-group deviation of a homo-oligomeric assembly
tm_score — TM-score of each structure against an external reference fold
tm_score_rna — TM-score_RNA (Gong 2019) against a reference fold via US-align, normalized by the reference length

Utility Blocks

annotate — Annotate molecular structures
filter — Filter pipeline rows by metrics predicate
ingest — Ingest data from various sources
materialize — Write stored structures to disk
mutate_residues — Create named protein sequence variants with explicit lineage
nanoparticle_prepare — Align antigen clash reference and preserve a C3 scaffold's symmetry axis
order_collection — Extract sequences, check ordered status, collect metrics
package_submission — Validate and write a sequence-only design submission
remap_ligand — Rename predicted ligand atom/residue names to match a reference CIF
restore_sidechains — Restore selected motif rotamers from an upstream structure
rotamer_relax — PyRosetta rotamer relaxation
rotamer_sweep — Enumerate explicit functional side-chain chi combinations
select_chains — Keep specified chains, preserving residue IDs and parent lineage
select_diverse — Select a diverse design subset under sequence/structure cutoffs
select_motif — Apply a function to select motif residues for downstream fixing
selective_sidechain_repack — Repack only explicitly selected protein sidechains
to_cif — Parse structures via get_data and re-save as CIF
to_pdb — Parse structures via get_data and re-save as PDB
transform — Apply an atomworks Transform to each pipeline row

Network Blocks¶

Blocks that run neural network inference — structure prediction, design, sequence models.

esmfold2¶

ESMFold2 structure prediction with optional MSA conditioning.

Class: ESMFold2Block · Input: True · Resources: partition=gpu, gpu_class=small, mem=48GB, time=5:00:00 · Container: esmfold2

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
`num_loops`	`int`	`3`	Trunk refinement loops (API default 3; reference driver used 20).
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`seed`	`Optional[int]`	`0`	Random seed for input prep + diffusion sampling.
`model`	`str`	`'biohub/ESMFold2'`	HuggingFace model id (or local snapshot path) for ESMFold2.
`use_local_msas`	`bool`	`False`	Condition each protein chain on its precomputed A3M from the on-disk MSA stores. False runs ESMFold2 in single-sequen...
`msa_dirs`	`Optional[list]`	`None`	Sharded MSA roots searched in order, as paths or {dir, extension, directory_depth} dicts. None uses PROTEIN_MSA_DIRS.
`max_msa_sequences`	`Optional[int]`	`16384`	Maximum A3M depth supplied to ESMFold2's MSA encoder.
`raise_if_missing_msa_for_protein_of_length_n`	`Optional[int]`	`None`	Fail when a protein chain at least this many residues long has no on-disk MSA instead of silently using single-sequen...

openfold3¶

OpenFold 3 biomolecular structure prediction.

Class: OpenFold3Block · Input: True · Resources: partition=gpu, gpu_class=small, mem=64GB, time=5:00:00 · Container: openfold3

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
`num_model_seeds`	`int`	`1`	Number of model seeds to use for each query.
`use_msa_server`	`bool`	`False`	Use ColabFold MSA server for alignments. False for single-sequence prediction.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`use_templates`	`bool`	`False`	Use ColabFold MSA server for template alignments.
`template_json`	`Optional[str]`	`None`	Path to template JSON file for OpenFold3 input. If provided, sequences from upstream designs are filled into this tem...
`cache_dir`	`str`	`<computed at runtime>`	Directory containing OpenFold3 model weights.
`runner_yaml`	`Optional[str]`	`None`	Path to a runner YAML for advanced model/dataset configuration.

boltz2¶

Boltz-2 biomolecular structure prediction.

Class: Boltz2Block · Input: True · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: boltz2

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
`override`	`bool`	`True`	Override existing predictions if found.
`step_scale`	`float`	`1.638`	Diffusion step scale. Lower = more diversity (recommended 1-2).
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`max_parallel_samples`	`int`	`1`	Max samples to predict in parallel.
`output_format`	`str`	`'mmcif'`	Output format: 'mmcif' or 'pdb'.
`use_msa_server`	`bool`	`False`	Auto-generate MSAs via mmseqs2 server.
`msa_server_url`	`Optional[str]`	`None`	MSA server URL (default: ColabFold).
`use_potentials`	`bool`	`False`	Use inference-time potentials for better physical plausibility.
`predict_affinity`	`bool`	`False`	Enable affinity prediction for protein-ligand complexes.
`cache_dir`	`str`	`<computed at runtime>`	Directory containing Boltz model weights and CCD data.
`template_yaml`	`Optional[str]`	`None`	Path to a template YAML file. If provided, sequences from upstream designs are filled into this template.
`no_kernels`	`bool`	`False`	Disable cuEquivariance kernels (for older GPUs).

boltz2_steered¶

Boltz-2 prediction with helix-steering from Protein Hunter.

Class: Boltz2SteeredBlock · Input: True · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: protein_hunter

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
`override`	`bool`	`True`	Override existing predictions if found.
`step_scale`	`float`	`1.638`	Diffusion step scale. Lower = more diversity (recommended 1-2).
`randomly_kill_helix_feature`	`bool`	`False`	Scale pair embeddings at (i, i+4) positions by a random negative factor on the binder chain, disrupting helix formation.
`negative_helix_constant`	`float`	`0.2`	Magnitude of the negative scaling factor for helix disruption. Higher = stronger suppression.
`binder_chain`	`str`	`'A'`	Chain ID of the binder to apply helix steering to.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`max_parallel_samples`	`int`	`1`	Max samples to predict in parallel.
`output_format`	`str`	`'mmcif'`	Output format: 'mmcif' or 'pdb'.
`use_msa_server`	`bool`	`False`	Auto-generate MSAs via mmseqs2 server.
`msa_server_url`	`Optional[str]`	`None`	MSA server URL (default: ColabFold).
`use_potentials`	`bool`	`False`	Use inference-time potentials for better physical plausibility.
`predict_affinity`	`bool`	`False`	Enable affinity prediction for protein-ligand complexes.
`cache_dir`	`str`	`<computed at runtime>`	Directory containing Boltz model weights and CCD data.
`template_yaml`	`Optional[str]`	`None`	Path to a template YAML file. If provided, sequences from upstream designs are filled into this template.
`no_kernels`	`bool`	`False`	Disable cuEquivariance kernels (for older GPUs).

rf3¶

Vanilla RosettaFold3 block for structure prediction (container-based).

Class: RosettaFold3Block · Input: True · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: foundry

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`10`	—
`num_diffusion_steps`	`int`	`50`	—
`ckpt_path`	`Optional[str]`	`<computed at runtime>`	Checkpoint path or Foundry registry name (default: rf3).
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`early_stopping_plddt_threshold`	`float`	`0.5`	Stop diffusion early if pLDDT falls below this threshold after the initial recycle. Set to 0.5 for 10-20x throughput ...
`dump_trajectories`	`bool`	`False`	Write the noisy and denoised diffusion trajectories for each sample. Trajectory files are large, place the final pred...
`seed_offset`	`int`	`42`	Base random seed; each chunk increments by 1. Ignored when model_seeds is set.
`model_seeds`	`Optional[list[int]]`	`None`	List of random seeds for multi-seed inference per input. Each seed runs a completely independent trunk pass, giving m...
`msa_dirs`	`Optional[list[str]]`	`None`	Absolute on-disk protein MSA roots. Chaperone exports these to both LOCAL_MSA_DIRS and PROTEIN_MSA_DIRS (supported Fo...
`raise_if_missing_msa_for_protein_of_length_n`	`Optional[int]`	`None`	Fail when a protein at least this many residues long has no MSA. Set to 0 to require an MSA for every protein chain.
`create_bond_graph`	`bool`	`False`	Assign ligand bonds from specified_smiles and write inputs as CIF with bond topology. Required when input structures ...
`input_mode`	`Literal`	`'file'`	'file' passes CIF/PDB directly (current default). 'sequence' builds RF3 component JSON from sequences + CCD/SMILES.
`template_json`	`Optional[str]`	`None`	Path to RF3 JSON template. Protein sequences are filled from upstream designs; everything else (ligands, bonds, templ...
`template_selection`	`Optional[list[str]]`	`None`	Token-level templates for polymers and small molecules. Uses AtomSelection format, e.g. ['A//30', 'A//53', 'X'].
`ground_truth_conformer_selection`	`Optional[list[str]]`	`None`	Lock selected residues/chains to ground truth conformers. Useful for keeping known ligand conformations while allowin...
`cyclic_chains`	`Optional[list[str]]`	`None`	Chain IDs that should be cyclized. If input is CIF, cyclization is auto-detected.

protenix¶

Protenix biomolecular structure prediction (v2).

Class: ProtenixBlock · Input: True · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: protenix

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
`model_name`	`str`	`'protenix_base_20250630_v1.0.0'`	Model checkpoint name. Options: protenix-v2 (464M, best accuracy, TFG support), protenix_base_20250630_v1.0.0 (368M, ...
`seeds`	`str`	`'101'`	Comma-separated random seeds for independent runs. Each seed runs a completely independent prediction. Example: '101,...
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`dtype`	`str`	`'bf16'`	Precision: 'bf16' (faster, default) or 'fp32' (more precise).
`use_msa`	`bool`	`False`	Enable multiple sequence alignment features. Requires HMMER databases accessible via /databases bind mount. Default F...
`use_template`	`bool`	`False`	Use template structures for guided prediction.
`use_tfg_guidance`	`bool`	`False`	Enable Training-Free Guidance (TFG) during diffusion sampling. TFG enforces geometric constraints (steric, torsion, b...
`need_atom_confidence`	`bool`	`False`	Output per-atom confidence scores (generates full_data JSON).
`enable_cache`	`bool`	`True`	Cache shareable diffusion variables for faster inference.
`enable_fusion`	`bool`	`True`	Enable kernel fusion optimizations.
`enable_tf32`	`bool`	`True`	Enable TF32 for FP32 matrix multiplications (faster on Ampere+).
`trimul_kernel`	`str`	`'cuequivariance'`	Triangle multiplication kernel: 'cuequivariance' or 'torch'.
`triatt_kernel`	`str`	`'cuequivariance'`	Triangle attention kernel: 'cuequivariance', 'deepspeed', 'triattention', or 'torch'.
`cache_dir`	`str`	`<computed at runtime>`	Directory for Protenix model weights (auto-downloaded on first run).
`template_json`	`Optional[str]`	`None`	Path to a template Protenix JSON file. If provided, protein sequences from upstream designs are filled into this temp...

chai1¶

Chai-1 biomolecular structure prediction.

Class: Chai1Block · Input: True · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: chai1

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`seed`	`Optional[int]`	`None`	Random seed for reproducibility. None = random.
`use_esm_embeddings`	`bool`	`True`	Use ESM embeddings for proteins.
`use_msa_server`	`bool`	`False`	Auto-generate MSAs via ColabFold mmseqs2 server.
`msa_server_url`	`Optional[str]`	`None`	MSA server URL (default: ColabFold).
`low_memory`	`bool`	`True`	Use memory-efficient inference mode.
`cache_dir`	`Optional[str]`	`<computed at runtime>`	Directory for Chai-1 model weights. Defaults to CHAI_DOWNLOADS_DIR setting.

af3¶

AlphaFold3 structure prediction.

Class: AlphaFold3Block · Input: True · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: alphafold3

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`jax_cache_dir`	`Optional[str]`	`<computed at runtime>`	Writable compilation cache mounted directly into the container
`input_mode`	`Literal`	`'mmcif'`	AF3 input preparation mode. The default 'mmcif' serializes the complete AtomArray and uses AF3's native Input.from_mm...
`use_local_msas`	`bool`	`False`	Inject each protein chain's precomputed a3m from the on-disk MSA stores into the AF3 input. Works with both native mm...
`msa_dirs`	`Optional[list]`	`None`	Sharded MSA roots searched in order, as paths or as {"dir", "extension", "directory_depth"} dicts. None uses PROTEIN_...
`max_msa_sequences`	`Optional[int]`	`10000`	Truncate injected MSAs to this depth.
`raise_if_missing_msa_for_protein_of_length_n`	`Optional[int]`	`None`	Fail when a protein at least this many residues long has no on-disk MSA, instead of silently falling back to single-s...
`template_json`	`Optional[str]`	`None`	Path to template JSON file for AF3 input. If not provided, create input from scratch.
`add_ptm_at_mapped_residues`	`Optional[dict]`	`None`	Map of {ptmType: original_residue_key} for covalent modifications. The residue key (e.g. 'A10') is looked up in the R...
`add_ptm_at_residues`	`Optional[list[dict]]`	`None`	Direct list of AF3 modification dicts to add to the protein sequence. Each entry should have 'ptmType' and 'ptmPositi...
`consolidate_intermediates`	`bool`	`True`	After compiling, pack each design's redundant per-sample subdirs (outputs//seed-_sample-/) into a single un...
`buckets`	`Optional[list[int]]`	`None`	Token-count buckets for JIT compilation. None uses the inner script's smart default (fine-grained at small token coun...
`flash_attention_implementation`	`str`	`'triton'`	'triton' (fastest, Ampere+), 'cudnn', or 'xla' (portable).
`conformer_max_iterations`	`Optional[int]`	`None`	RDKit conformer search max iterations. None uses RDKit default.
`mean_pae_contact_distance`	`float`	`5.0`	Contact cutoff (Å) for the `interface_mean_pae` metric: only protein residues with any atom within this distance of t...
`mean_pae_core_atoms`	`Optional[list[str]]`	`None`	Ligand atom names defining the buried/core substructure for the `interface_mean_pae` and `full_mean_pae` metrics. Whe...
`mean_pae_sidechain_angle`	`Optional[float]`	`90.0`	Max angle (degrees) between a pocket residue's CA→sidechain-centroid vector and its CA→nearest-ligand-atom vector for...
`model_seeds`	`list[int]`	`<computed at runtime>`	Independent AF3 model seeds written into each input JSON. Multiple seeds rerun trunk and diffusion inference and are ...

af3_with_target_template¶

AF3 with the design's target chain attached as a structural template.

Class: AF3WithTargetTemplate · Input: True · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: alphafold3

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`jax_cache_dir`	`Optional[str]`	`<computed at runtime>`	Writable compilation cache mounted directly into the container
`input_mode`	`Literal`	`'json'`	—
`use_local_msas`	`bool`	`False`	Inject each protein chain's precomputed a3m from the on-disk MSA stores into the AF3 input. Works with both native mm...
`msa_dirs`	`Optional[list]`	`None`	Sharded MSA roots searched in order, as paths or as {"dir", "extension", "directory_depth"} dicts. None uses PROTEIN_...
`max_msa_sequences`	`Optional[int]`	`10000`	Truncate injected MSAs to this depth.
`raise_if_missing_msa_for_protein_of_length_n`	`Optional[int]`	`None`	Fail when a protein at least this many residues long has no on-disk MSA, instead of silently falling back to single-s...
`template_json`	`Optional[str]`	`None`	Path to template JSON file for AF3 input. If not provided, create input from scratch.
`add_ptm_at_mapped_residues`	`Optional[dict]`	`None`	Map of {ptmType: original_residue_key} for covalent modifications. The residue key (e.g. 'A10') is looked up in the R...
`add_ptm_at_residues`	`Optional[list[dict]]`	`None`	Direct list of AF3 modification dicts to add to the protein sequence. Each entry should have 'ptmType' and 'ptmPositi...
`consolidate_intermediates`	`bool`	`True`	After compiling, pack each design's redundant per-sample subdirs (outputs//seed-_sample-/) into a single un...
`buckets`	`Optional[list[int]]`	`None`	Token-count buckets for JIT compilation. None uses the inner script's smart default (fine-grained at small token coun...
`flash_attention_implementation`	`str`	`'triton'`	'triton' (fastest, Ampere+), 'cudnn', or 'xla' (portable).
`conformer_max_iterations`	`Optional[int]`	`None`	RDKit conformer search max iterations. None uses RDKit default.
`mean_pae_contact_distance`	`float`	`5.0`	Contact cutoff (Å) for the `interface_mean_pae` metric: only protein residues with any atom within this distance of t...
`mean_pae_core_atoms`	`Optional[list[str]]`	`None`	Ligand atom names defining the buried/core substructure for the `interface_mean_pae` and `full_mean_pae` metrics. Whe...
`mean_pae_sidechain_angle`	`Optional[float]`	`90.0`	Max angle (degrees) between a pocket residue's CA→sidechain-centroid vector and its CA→nearest-ligand-atom vector for...
`model_seeds`	`list[int]`	`<computed at runtime>`	Independent AF3 model seeds written into each input JSON. Multiple seeds rerun trunk and diffusion inference and are ...
`target_chain`	`str`	`'B'`	Chain ID of the receptor / target whose structure should be templated.
`template_kalign_matches`	`Optional[int]`	`None`	If set, override AF3's `templates[].kalign_matches`. Default lets AF3 decide.
`full_receptor_pdb`	`Optional[str]`	`None`	Optional path to a PDB/CIF containing the full target chain. When set, AF3's chain `target_chain` is replaced with th...

colabfold¶

ColabFold (AlphaFold2) monomer and multimer structure prediction.

Class: ColabFoldBlock · Input: True · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: colabfold_portable

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`5`	Number of diffusion samples (prediction structures) per input.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`num_diffusion_steps`	`int`	`200`	Number of diffusion sampling steps.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`num_models`	`int`	`5`	Number of AF2 models (1-5).
`num_seeds`	`int`	`1`	Number of random seeds per model.
`random_seed`	`Optional[int]`	`None`	Base random seed. None = ColabFold default (0).
`msa_mode`	`str`	`'single_sequence'`	MSA databases: single_sequence (no MSA, offline), mmseqs2_uniref_env, mmseqs2_uniref, mmseqs2_uniref_env_envpair.
`use_msa_server`	`bool`	`False`	Query the ColabFold MMseqs2 server for MSAs. When True and msa_mode is still single_sequence, switches to mmseqs2_uni...
`use_local_msas`	`bool`	`False`	Look each folded sequence up in the precomputed on-disk MSA stores and stage it as an a3m input, which overrides msa_...
`msa_dirs`	`Optional[list]`	`None`	Sharded MSA roots searched in order, as paths or as {"dir", "extension", "directory_depth"} dicts. None uses the site...
`max_msa_sequences`	`Optional[int]`	`10000`	Truncate staged MSAs to this depth. None keeps the full store entry.
`raise_if_missing_msa_for_protein_of_length_n`	`Optional[int]`	`None`	Fail when a protein at least this many residues long has no on-disk MSA. None (default) uses a query-only alignment. ...
`model_type`	`str`	`'auto'`	AF2 model set (auto, alphafold2_ptm, alphafold2_multimer_v3, ...).
`use_templates`	`bool`	`False`	Query PDB templates from the MSA server (needs internet).
`amber_relax`	`bool`	`False`	Run AMBER relaxation (CPU only; --use-gpu-relax is never emitted).
`rank`	`str`	`'auto'`	Model ranking metric (auto, plddt, ptm, iptm, multimer).
`data_dir`	`Optional[str]`	`<computed at runtime>`	Writable dir for AF2 params (--data). ~3-4GB downloaded on first run. Bound into the container automatically when the...

bindcraft¶

Design protein binders against each input target structure.

Class: BindCraftBlock · Input: True · Resources: partition=gpu, gpu_class=large, mem=48GB, time=24:00:00 · Container: bindcraft

Parameter	Type	Default	Description
`chains`	`str`	required	BindCraft target-chain selector, e.g. 'A' or 'A,B'.
`target_hotspot_residues`	`Optional[str]`	`None`	Hotspot selection such as 'A56,A60-64'; None auto-selects a site.
`lengths`	`tuple`	`(65, 150)`	Inclusive minimum and maximum binder length.
`number_of_final_designs`	`int`	`1`	Accepted designs requested per input target.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`max_trajectories`	`Optional[int]`	`100`	Maximum hallucination trajectories per target. None honors the advanced preset and may run until the accepted-design ...
`binder_name`	`str`	`'binder'`	Prefix for generated designs.
`filter_preset`	`str`	`'default_filters'`	Bundled BindCraft filter preset basename.
`advanced_preset`	`str`	`'default_4stage_multimer'`	Bundled BindCraft advanced-settings preset basename.
`filters_path`	`Optional[str]`	`None`	Optional custom filters JSON; overrides filter_preset.
`advanced_path`	`Optional[str]`	`None`	Optional custom advanced JSON; overrides advanced_preset.
`filter_overrides`	`dict`	`<computed at runtime>`	Shallow overrides applied to the selected filter JSON.
`advanced_overrides`	`dict`	`<computed at runtime>`	Shallow overrides applied to the selected advanced JSON.

bindcraft2¶

Design miniprotein binders to one protein chain and return accepted refolds.

Inputs and outputs: Installation and binder guide. One protein target chain per input; returns accepted, refolded target–binder complexes with side chains and native scores. Finite trajectory budgets fail if too few designs pass. Uses its own JAX/CUDA 12 runtime and upstream hosting-restricted license.

Class: BindCraft2Block · Input: True · Resources: partition=gpu, gpu_class=large, mem=64GB, time=24:00:00 · Container: bindcraft2

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`chain`	`str`	required	Single target protein chain in the input structure
`hotspots`	`Optional[str]`	`None`	Target residue numbers/ranges, e.g. 54,56,66-70
`lengths`	`tuple`	`(60, 180)`	Inclusive binder length bounds
`number_of_final_designs`	`int`	`1`	—
`max_trajectories`	`int`	`10`	Finite per-input optimization budget; failure if insufficient designs pass
`seed`	`int`	`42`	—
`af2_params_dir`	`Optional[str]`	`None`	Directory containing all five multimer-v3 and two PTM AF2 checkpoints; alternatively BINDCRAFT_AF2_PARAMS
`weights_dir`	`str`	`<computed at runtime>`	Writable upstream weight download cache when no AF2 directory is supplied
`overrides`	`dict`	`<computed at runtime>`	Selected upstream scientific settings; changing filters changes the acceptance criterion

boltzgen¶

BoltzGen block for protein design.

Class: BoltzGenBlock · Input: False · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: boltzgen

Parameter	Type	Default	Description
`config`	`str`	required	Path to BoltzGen design specification YAML file
`protocol`	`str`	`'protein-anything'`	Design protocol (protein-anything, peptide-anything, protein-small_molecule, nanobody-anything, antibody-anything, pr...
`num_designs`	`int`	`100`	Total number of designs to generate
`budget`	`int`	`30`	Number of final designs after filtering
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`steps`	`Optional[list[str]]`	`None`	Pipeline steps to run (default: all). Options: design, inverse_folding, design_folding, folding, affinity, analysis, ...
`diffusion_batch_size`	`Optional[int]`	`None`	Designs per trunk run (default: auto)
`step_scale`	`Optional[float]`	`None`	Fixed diffusion step scale
`noise_scale`	`Optional[float]`	`None`	Fixed noise scale
`skip_inverse_folding`	`bool`	`False`	Skip the inverse folding step
`inverse_fold_num_sequences`	`int`	`1`	Number of sequences per backbone
`alpha`	`Optional[float]`	`None`	Quality vs diversity (0.0=quality, 1.0=diversity)
`devices`	`Optional[int]`	`None`	Number of GPUs (default: all available)
`reuse`	`bool`	`False`	Reuse existing results from previous runs
`n_batches`	`int`	`1`	Number of independent BoltzGen runs

disco¶

DISCO: Diffusion for Sequence-Structure Co-design.

Class: DISCOBlock · Input: False · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: disco

Parameter	Type	Default	Description
`inputs`	`str`	required	Path to DISCO input JSON file (AF3 Server-like format)
`num_seeds`	`int`	`5`	Number of random seeds (samples per job in the input JSON)
`experiment`	`Literal`	`'designable'`	Experiment preset: 'designable' (higher designability) or 'diverse' (greater structural variety)
`effort`	`Literal`	`'fast'`	Compute preset: 'fast' (~4x faster, default) or 'max' (full quality). Use 'max' for conditional generation.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_batches`	`int`	`1`	Number of parallel chunks to split seeds across
`use_deepspeed_evo_attention`	`bool`	`True`	Use DeepSpeed EvoformerAttention (requires CUTLASS + Ampere+ GPU)
`n_diffusion_steps`	`Optional[int]`	`None`	Override number of diffusion steps (default: set by effort preset)
`n_cycles`	`Optional[int]`	`None`	Override number of recycling cycles (default: set by effort preset)
`noisy_guidance`	`Optional[bool]`	`None`	Enable/disable noisy guidance (default: set by experiment preset)
`ckpt_path`	`Optional[str]`	`None`	Path to custom model checkpoint (default: auto-download from HuggingFace)
`cache_dir`	`str`	`<computed at runtime>`	Hugging Face model cache mounted into the container

e1¶

E1 block for masked language modeling on protein sequences.

Class: E1Block · Input: True · Resources: partition=gpu, gpu_class=small, mem=16GB, time=5:00:00 · Container: e1

Parameter	Type	Default	Description
`model_name`	`Literal \\| str`	`'150m'`	Model size (150m, 300m, 600m) or full HuggingFace path
`max_batch_tokens`	`int`	`65536`	Maximum number of tokens to batch in a single forward pass
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`save_masked_positions_only`	`bool`	`False`	Whether to save logits/embeddings for only the masked positions
`fields_to_save`	`list[str]`	`['logits', 'token_embeddings', 'mean_token_embeddings']`	Fields to save in output safetensors
`context_fasta`	`Optional[str]`	`None`	Optional path to context FASTA for retrieval-augmented prediction
`cache_dir`	`str`	`<computed at runtime>`	Hugging Face model cache, mounted directly into the container

esm2¶

ESM-2 block for protein sequence embeddings.

Class: ESM2Block · Input: True · Resources: partition=gpu, gpu_class=small, mem=16GB, time=5:00:00 · Container: esm2

Parameter	Type	Default	Description
`model_name`	`Literal \\| str`	`'650m'`	Model size (8m, 35m, 150m, 650m, 3b) or full HuggingFace path
`max_batch_tokens`	`int`	`4096`	Maximum number of tokens to batch in a single forward pass
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`fields_to_save`	`list[str]`	`['token_embeddings', 'mean_token_embeddings']`	Fields to save in output safetensors

foundry_mpnn¶

Foundry Protein/LigandMPNN for inverse folding.

Validation status: Wrapper/integration coverage does not establish scientific equivalence to the mpnn block (LigandMPNN). The PPI benchmark scripts provide a starting point for a controlled comparison; no equivalence result is asserted here.

Class: FoundryMPNNBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00 · Container: foundry

Parameter	Type	Default	Description
`model_type`	`Literal`	`'ligand_mpnn'`	Model type: protein_mpnn or ligand_mpnn.
`num_seq_per_pdb`	`int`	`8`	Number of sequences to generate per structure.
`output_mode`	`Literal`	`'sequences'`	Return compact sequence rows or legacy generated CIF structures.
`temperature`	`float`	`0.1`	Sampling temperature. Lower = more conservative.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`checkpoint_path`	`Optional[str]`	`None`	Path to model checkpoint. None uses foundry default.
`seed`	`Optional[int]`	`None`	Random seed for reproducibility.
`fixed_residues`	`Optional[list[str]]`	`None`	Residues to keep fixed, e.g. ['A12', 'A13'].
`designed_residues`	`Optional[list[str]]`	`None`	Residues to redesign (all others fixed).
`fixed_chains`	`Optional[list[str]]`	`None`	Chain IDs to keep fixed.
`designed_chains`	`Optional[list[str]]`	`None`	Chain IDs to redesign.
`keep_motif_sequences_fixed`	`bool`	`True`	Keep sequences of motifs fixed (from RFD3 metadata).

manifest_val_structures¶

Materialize the parquet's PDB assemblies and carry its scoring schema.

Class: ManifestValStructuresBlock · Input: False · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`parquet_path`	`str`	required	Entry-level validation table (.parquet or .csv).
`pdb_mirror`	`str`	`<computed at runtime>`	Local PDB mirror root, sharded by the middle two PDB-ID characters.
`bucket`	`str`	`'all'`	Validation composition bucket to select. 'all' keeps every row; individual buckets are: protein, protein-protein, pro...

mpnn¶

Inverse folding with LigandMPNN.

Backend: mpnn runs LigandMPNN (dauparas/LigandMPNN) from the ligandmpnn image, with every official checkpoint baked in; install it with chaperone install ligandmpnn. No settings are needed. model_type selects the model (protein_mpnn, ligand_mpnn, soluble_mpnn, per_residue_label_membrane_mpnn or global_label_membrane_mpnn) and noise_level selects the checkpoint version; the container guide lists the valid combinations.

Packed structures: Use output_mode="structures" and, for one structure per sampled sequence, args={"number_of_packs_per_design": 1}. The worked examples set these explicitly. The block default remains compact sequence output. With packed output and args.repack_everything=0, Chaperone restores observed coordinates of fixed protein residues after packing: the upstream packer can reconstruct and move these atoms even when their sequences are fixed. Newly completed atoms remain packed. Set args.repack_everything=1 to allow fixed residues to be repacked, or preserve_fixed_coordinates=False to retain native packer reconstructions. Sequence fixing and coordinate preservation are distinct from subsequent refolding, which can move the structure.

Class: ClassicMPNNBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00 · Container: ligandmpnn

Parameter	Type	Default	Description
`model_type`	`Literal`	`'ligand_mpnn'`	Model family: protein_mpnn, ligand_mpnn, soluble_mpnn, or a membrane model (per_residue_label_membrane_mpnn, global_l...
`noise_level`	`Optional[Literal]`	`None`	Checkpoint version: backbone noise (Angstrom) the model was trained with. None uses the upstream default. Valid value...
`num_seq_per_pdb`	`int`	`8`	Number of sequences to generate per structure
`output_mode`	`Literal`	`'sequences'`	Return compact sequence rows (default) or side-chain-packed structures. Sequence mode disables packing.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`preserve_fixed_coordinates`	`bool`	`True`	For packed structures with repack_everything=0, restore observed finite heavy-atom coordinates of fixed protein resid...
`skip_existing`	`bool`	`True`	Skip processing if outputs already exist
`device`	`Optional[str]`	`None`	Force device for MPNN inference (e.g. 'cpu'). Default: auto-detect.
`keep_motif_sequences_fixed`	`bool`	`True`	Keep sequences of motifs fixed (if read into the dataframe)
`fix_interface_residues`	`bool`	`False`	Fix the sequence at the interfaces
`hbond_fix`	`bool`	`False`	Fix residues involved in hydrogen bonds
`aromatic_fix`	`bool`	`False`	Fix aromatic residues near ligand
`component_remap`	`Optional[dict]`	`None`	Remap residue names in CIF files before MPNN. Maps old -> new (e.g., {'LIG_B': 'B;N'}).
`args`	`MPNNArgs`	`<computed at runtime>`	LigandMPNN/ProteinMPNN CLI arguments. Partial overrides keep defaults.

protein_hunter¶

Protein Hunter: de novo protein binder design via structure hallucination with Boltz.

Class: ProteinHunterBlock · Input: False · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: protein_hunter

Parameter	Type	Default	Description
`protein_seqs`	`str`	`''`	Target protein sequence(s). Separate multiple chains with ':'.
`num_designs`	`int`	`10`	Number of independent design trajectories.
`num_cycles`	`int`	`5`	Number of predict-redesign cycles per trajectory.
`min_protein_length`	`int`	`100`	Minimum binder length (residues).
`max_protein_length`	`int`	`150`	Maximum binder length (residues).
`diffuse_steps`	`int`	`200`	Number of diffusion steps.
`recycling_steps`	`int`	`3`	Number of recycling steps.
`temperature`	`float`	`0.1`	LigandMPNN sampling temperature.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`ligand_ccd`	`str`	`''`	CCD code for small molecule target (e.g., 'SAM').
`ligand_smiles`	`str`	`''`	SMILES string for small molecule target.
`nucleic_seq`	`str`	`''`	DNA/RNA target sequence.
`nucleic_type`	`str`	`'dna'`	Nucleic acid type: 'dna' or 'rna'.
`emit_all_cycles`	`bool`	`False`	Emit one handler row per predicted cycle instead of one multi-structure lineage row per trajectory. Use this when eve...
`percent_X`	`int`	`90`	Percentage of X (unknown) residues in initial sequence.
`seq`	`str`	`''`	Starting binder sequence (for refiner mode). Empty = random.
`refiner_mode`	`bool`	`False`	Refine the sequence supplied by seq instead of initializing a de novo binder trajectory.
`msa_mode`	`str`	`'mmseqs'`	MSA mode: 'mmseqs' or 'single'.
`boltz_model_version`	`str`	`'boltz2'`	Boltz model version.
`high_iptm_threshold`	`float`	`0.8`	ipTM threshold for saving designs.
`high_plddt_threshold`	`float`	`0.8`	pLDDT threshold for saving designs.
`cyclic`	`bool`	`False`	Enable cyclic peptide design.
`alanine_bias`	`bool`	`False`	Apply negative alanine bias during design.
`randomly_kill_helix_feature`	`bool`	`False`	Disrupt binder i/i+4 pair features during diffusion to reduce alpha-helical collapse and explore alternative topologies.
`negative_helix_constant`	`float`	`0.2`	Magnitude of the random negative i/i+4 pair-feature scaling used when randomly_kill_helix_feature is enabled.
`omit_AA`	`str`	`'C'`	Amino acids to omit from design (e.g., 'C').
`contact_residues`	`str`	`''`	Target chain residue positions for contact specification. Comma-separated, pipe-separated for multiple chains (e.g., ...
`template_path`	`str`	`''`	PDB code or path(s) for template conditioning.
`template_cif_chain_id`	`str`	`''`	Chain ID(s) for mmCIF templates.
`ccd_path`	`str`	`<computed at runtime>`	Path to CCD molecule data directory (.pkl files) for Boltz.
`boltz_model_path`	`str`	`<computed at runtime>`	Path to Boltz model checkpoint.
`gpu_id`	`int`	`0`	GPU device ID.

proteina_complexa¶

Co-design all-atom protein binders around a fixed protein target.

Inputs and outputs: Binder configuration guide. Single-pass all-atom protein binder generation from a target PDB, target selection and hotspots. Outputs include target and designed binder. Run a downstream structure predictor to evaluate binding geometry and confidence; generation alone is not a binding-quality result.

Class: ProteinaComplexaBlock · Input: False · Resources: partition=gpu, gpu_class=large, mem=48GB, time=5:00:00 · Container: proteina_complexa

Parameter	Type	Default	Description
`target_pdb`	`str`	required	Target PDB; copied into each job sandbox
`target_selection`	`str`	required	Upstream target input specification, e.g. A1-115
`hotspots`	`list[str]`	required	Target residue identifiers, e.g. ['A37', 'A39']
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`binder_length`	`tuple`	`(80, 120)`	Inclusive minimum and maximum binder residue count
`num_designs`	`int`	`1`	Designs generated per batch
`n_batches`	`int`	`1`	Independent engine jobs, each with a distinct seed
`batch_size`	`int`	`1`	Simultaneous samples on one GPU
`num_steps`	`int`	`400`	Flow integration steps; upstream default is 400
`seed`	`int`	`0`	Base seed, incremented by engine chunk ID
`checkpoint_dir`	`str`	required	Directory containing public complexa.ckpt and complexa_ae.ckpt checkpoints

rfd1¶

RFdiffusion v1 block for backbone-level protein design.

Class: RFD1Block · Input: False · Resources: partition=gpu, gpu_class=small, mem=16GB, time=5:00:00 · Container: rfdiffusion1

Parameter	Type	Default	Description
`contigs`	`str`	required	Contig specification. Segments separated by /, chains by spaces. E.g. '150-150' for unconditional, '5-15/A10-25/30-40...
`num_designs`	`int`	`10`	Number of designs to generate per chunk
`input_pdb`	`Optional[str]`	`None`	Input PDB for motif scaffolding or binder design
`config_name`	`str`	`'base'`	Hydra config name (e.g. 'base', 'symmetry')
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`symmetry`	`Optional[str]`	`None`	Symmetry type for oligomer design (e.g. 'c3', 'c5', 'd3', 'tetrahedral'). Sets config_name='symmetry' automatically.
`diffuser_T`	`int`	`50`	Number of denoising timesteps
`partial_T`	`Optional[int]`	`None`	Partial diffusion timesteps (for design diversification)
`noise_scale_ca`	`Optional[float]`	`None`	Translation noise scaling (denoiser.noise_scale_ca)
`noise_scale_frame`	`Optional[float]`	`None`	Rotation noise scaling (denoiser.noise_scale_frame)
`ckpt_override_path`	`Optional[str]`	`None`	Custom checkpoint path. Defaults to container's Base_ckpt.pt. Common options: Complex_base_ckpt.pt, Complex_Fold_base...
`hotspot_res`	`Optional[list[str]]`	`None`	Hotspot residues for binder design, e.g. ['A30', 'A33', 'A34']
`inpaint_seq`	`Optional[list[str]]`	`None`	Inpaint sequence positions, e.g. ['A1', 'A30-40']
`inpaint_str`	`Optional[list[str]]`	`None`	Inpaint structure positions (flexible regions), e.g. ['B165-178']
`scaffold_guided`	`bool`	`False`	Enable fold-conditioned design with ss/adj constraints
`target_pdb`	`Optional[str]`	`None`	Target PDB for scaffold-guided binder design
`scaffold_dir`	`Optional[str]`	`None`	Directory of _ss.pt and _adj.pt files for scaffold guidance
`extra_hydra_args`	`Optional[list[str]]`	`None`	Additional Hydra CLI overrides passed verbatim, e.g. ['diffuser.chi_type=interp']
`raise_on_missing_trbs`	`bool`	`True`	Raise error if TRB file is missing for a design

rfd2¶

RFdiffusion2 (rf_diffusion) block for atomic-level protein design.

Class: RFD2Block · Input: False · Resources: partition=gpu, gpu_class=small, mem=32GB, time=5:00:00 · Container: rfdiffusion2_portable

Parameter	Type	Default	Description
`inputs`	`str \\| list[str]`	required	Hydra config name to use (references a yaml/json file in benchmark configs)
`num_designs`	`int`	`1`	Number of designs to generate per benchmark case
`ckpt_path`	`str`	`<computed at runtime>`	Path to RFdiffusion2 checkpoint file. The portable image defaults to the RFD_173.pt baked into it.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`raise_on_missing_trbs`	`bool`	`True`	Raise error if TRB file is missing for a design

rfd3¶

RFdiffusion3 block for backbone generation.

Class: RFD3Block · Input: False · Resources: partition=gpu, gpu_class=small, mem=16GB, time=5:00:00 · Container: foundry

Parameter	Type	Default	Description
`inputs`	`str \\| dict \\| None`	`None`	Path to input json/yaml file, or dict of design specifications. JSON inputs take the following nested top-level struc...
`ckpt_path`	`Optional[str]`	`<computed at runtime>`	Optional checkpoint path (foundry will manage checkpoints itself)
`batch_size`	`int`	`8`	Number of designs to diffuse per batch
`n_batches`	`int`	`2`	Number of diffusion batches to run (in total)
`seed_offset`	`Optional[int]`	`42`	Increments seed by 1 from offset for every chunk
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`compile_model`	`Optional[bool]`	`None`	Compile hot diffusion modules; None inherits the Foundry runtime default.
`json_keys_subset`	`list`	`<computed at runtime>`	Keys to fetch from json
`jitter_ori_sigma`	`float`	`0`	Jitter origin of the noise cloud by this standard deviation
`convert_config_to_test_config`	`bool`	`False`	When True, shorten generated protein length to 50 residues and use fast inference settings for testing. Only shortens...
`inference_sampler`	`dict`	`<computed at runtime>`	Sampler parameters to override (not configurable via CLI)

rfd3_partial¶

RFdiffusion3 partial diffusion. Useful for diversifying structures.

Class: RFD3PartialBlock · Input: True · Resources: partition=gpu, gpu_class=small, mem=16GB, time=5:00:00 · Container: foundry

Parameter	Type	Default	Description
`inputs`	`str \\| dict \\| None`	`None`	Path to input json/yaml file, or dict of design specifications. JSON inputs take the following nested top-level struc...
`ckpt_path`	`Optional[str]`	`<computed at runtime>`	Optional checkpoint path (foundry will manage checkpoints itself)
`batch_size`	`int`	`8`	Number of designs to diffuse per batch
`n_batches`	`int`	`2`	Number of diffusion batches to run (in total)
`seed_offset`	`Optional[int]`	`42`	Increments seed by 1 from offset for every chunk
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`compile_model`	`Optional[bool]`	`None`	Compile hot diffusion modules; None inherits the Foundry runtime default.
`json_keys_subset`	`list`	`<computed at runtime>`	Keys to fetch from json
`jitter_ori_sigma`	`float`	`0`	Jitter origin of the noise cloud by this standard deviation
`convert_config_to_test_config`	`bool`	`False`	When True, shorten generated protein length to 50 residues and use fast inference settings for testing. Only shortens...
`inference_sampler`	`dict`	`<computed at runtime>`	Sampler parameters to override (not configurable via CLI)
`partial_t`	`int`	`5`	Timestep for partial diffusion (e.g., 5 = ~80% diffused).
`partial_diffusion_mode`	`str`	`'indexed'`	Mode for partial diffusion
`ligand_name`	`Optional[str]`	`None`	Ligand name for partial diffusion specs (e.g., 'LIG_B'). Overrides ligand from metadata when set.

rfd3na¶

RFdiffusion3NA block for nucleic-acid and mixed-polymer design.

Class: RFD3NABlock · Input: False · Resources: partition=gpu, gpu_class=small, mem=32GB, time=6:00:00 · Container: foundry

Parameter	Type	Default	Description
`inputs`	`str \\| dict \\| None`	`None`	Path to input json/yaml file, or dict of design specifications. JSON inputs take the following nested top-level struc...
`ckpt_path`	`Optional[str]`	`<computed at runtime>`	RFD3NA checkpoint path. Set to 'rfd3na' or None to use Foundry's checkpoint registry/search path.
`batch_size`	`int`	`8`	Number of designs to diffuse per batch
`n_batches`	`int`	`2`	Number of diffusion batches to run (in total)
`seed_offset`	`Optional[int]`	`42`	Increments seed by 1 from offset for every chunk
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`compile_model`	`Optional[bool]`	`None`	Compile hot diffusion modules; None inherits the Foundry runtime default.
`json_keys_subset`	`list`	`<computed at runtime>`	Keys to fetch from json
`jitter_ori_sigma`	`float`	`0`	Jitter origin of the noise cloud by this standard deviation
`convert_config_to_test_config`	`bool`	`False`	When True, shorten generated protein length to 50 residues and use fast inference settings for testing. Only shortens...
`inference_sampler`	`dict`	`<computed at runtime>`	RFD3NA inference-sampler overrides.
`read_sequence_from_sequence_head`	`bool`	`False`	Upstream README recommends False as the global setting for RFD3NA.
`prevalidate_inputs`	`bool`	`True`	Validate every RFD3NA input specification before loading the model.
`low_memory_mode`	`bool`	`False`	Use memory-efficient tokenization when GPU memory is constrained.
`dump_trajectories`	`bool`	`False`	Write large noisy and denoised diffusion trajectory files.
`skip_existing`	`bool`	`True`	Skip designs whose output file already exists.

rfoptimization¶

Refine chain A of an existing complex with RF3/Boltz and MPNN.

Inputs and outputs: Configuration guide. Refines an existing complex with designed protein chain A. The final MPNN redesign has incomplete side chains at mutations; pack/refold it before evaluating that sequence. Intermediate prediction confidence is retained in the upstream records, not attached as a score for the final redesigned structure.

Class: RFOptimizationBlock · Input: True · Resources: partition=gpu, gpu_class=large, mem=64GB, time=04:00:00 · Container: rfoptimization

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`total_cycles`	`int`	`10`	Number of RF3/Boltz + MPNN cycles per input.
`backprop_fraction`	`float`	`0.5`	Probability of RF3 gradients per cycle; otherwise Boltz-2.
`mpnn_model_type`	`Literal`	`'protein_mpnn'`	—
`mpnn_temperature`	`float`	`0.1`	—
`seed`	`int`	`42`	—
`rf3_checkpoint`	`Optional[str]`	`None`	Public confidence-enabled RF3 checkpoint; required when RF3 is used. Defaults to configured Foundry checkpoint direct...
`boltz_cache`	`str`	`<computed at runtime>`	—
`rf3_config`	`Optional[str]`	`None`	Optional upstream optimizer YAML. Omit to use the pinned cycle_ppi preset.
`msa_paths`	`dict`	`<computed at runtime>`	Optional chain to A3M path mapping; no remote MSA service is used.
`templates`	`dict`	`<computed at runtime>`	Optional chain to template path mapping for the RF3 branch.
`cyclic`	`bool`	`False`	Mark chain A cyclic in Boltz only; RF3 does not constrain its closure.

rpxdock¶

Dock a z-oriented C3 asymmetric unit into an I3 cage and emit C3 trimers

Input geometry must use the C3 symmetry axis as its z axis and retain the asymmetric unit's radial displacement. Outputs retain motif correspondence, input lineage and each model's RPX score. Changing scoring residue selection requires checking score thresholds; the score alone does not validate a complete cage or experimental assembly.

Class: RPXDockBlock · Input: True · Resources: partition=cpu, mem=8GB, time=02:00:00 · Container: rpxdock

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`nout_top`	`int`	`1`	Number of highest scoring C3 trimers per input
`beam_size`	`int`	`10000`	RPXdock hierarchical search beam size
`ignored_aas`	`str`	`'CGP'`	Residue types excluded from RPX scoring; use an empty string for glycine-placeholder backbones.

score_val_structures¶

Score the last (predicted) structure against the first (reference) one.

Class: ScoreValStructuresBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`metrics`	`list[ValidationMetric]`	`<computed at runtime>`	Scoring backends to run: rf3_allatom_lddt and/or peppr.
`remap_outputs_to_input`	`bool`	`False`	Map a renumbered monomer prediction onto the single scored protein pn-unit before scoring. Disabled by default so dir...
`min_atom_coverage`	`float`	`0.95`	Minimum fraction of resolved reference atoms that a remapped oracle output must represent. Used only when remap_outpu...

simplefold¶

Apple SimpleFold single-sequence, protein-monomer structure prediction.

Class: SimpleFoldBlock · Input: True · Resources: partition=gpu, gpu_class=large, mem=96GB, time=24:00:00 · Container: simplefold

Parameter	Type	Default	Description
`diffusion_batch_size`	`int`	`1`	Number of independently sampled conformers per protein.
`read_most_confident_model`	`bool`	`True`	If True, read only the highest-confidence model; False, read all samples.
`num_diffusion_steps`	`int`	`500`	Number of Euler-Maruyama flow-sampling steps.
`model`	`Literal`	`'simplefold_100M'`	Released SimpleFold folding checkpoint.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`n_recycles`	`Optional[int]`	`None`	Number of trunk recycling steps. None uses the model default.
`ckpt_path`	`Optional[str]`	`None`	Optional checkpoint path for model weights.
`compile_as_metric`	`bool`	`False`	If True, compile outputs as metrics with path_handle instead of path.
`remove_ligands`	`bool`	`False`	If True, skip ligands in input preparation.
`specified_smiles`	`dict`	`<computed at runtime>`	Map ligand name → SMILES string.
`specified_ccd`	`dict`	`<computed at runtime>`	Map ligand name → CCD code.
`disable_ccd_detection`	`bool`	`False`	If True, disable automatic CCD code detection.
`which`	`list[int] \\| str`	`'last'`	Which structure(s) in each row's lineage to predict. 'last' (default) uses the most recent; a list of ints selects by...
`sort_by_length`	`bool`	`True`	If True (default), sort the input handler by `sort_length_key` before chunking so each chunk hits a narrow JIT-bucket...
`sort_length_key`	`str`	`'rfd3.num_residues'`	Metric column to sort on when sort_by_length=True (dot-path into metrics JSON, as exposed by handler.unpack()). If mi...
`tau`	`float`	`0.01`	Diffusion coefficient scaling factor.
`seed`	`int`	`42`	Random seed; chunk id is added to keep chunks independent.
`predict_plddt`	`bool`	`False`	Run the optional pLDDT head. This additionally loads a fixed 1.6B folding model and is substantially more memory inte...
`raw_output_format`	`Literal`	`'pdb'`	SimpleFold's raw output format before compile-time conversion.
`convert_outputs_to_cif`	`bool`	`True`	Parse raw predictions into AtomArrays and write canonical CIF outputs.
`weights_dir`	`str`	`<computed at runtime>`	Shared directory for folding, ESM2, and CCD downloads.

tmol_interface¶

Protein<->ligand interface energetics (single-point, beta_nov2016).

Class: TmolInterfaceBlock · Input: True · Resources: partition=cpu, mem=16GB, time=5:00:00 · Container: tmol

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`ligand_smiles`	`Optional[str]`	`None`	SMILES of the ligand. Used to generate a tmol .tmol params file (typing only — input ligand coordinates are scored as...
`ligand_params_files`	`Optional[list[str]]`	`None`	Pre-made tmol .tmol params file(s) for the ligand(s). Most deterministic typing route; bypasses SMILES generation.

tmol_minimize¶

Cartesian energy minimisation (beta_nov2016_cart), all atoms free.

Class: TmolMinimizeBlock · Input: True · Resources: partition=cpu, mem=16GB, time=5:00:00 · Container: tmol

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`ligand_smiles`	`Optional[str]`	`None`	SMILES of the ligand. Used to generate a tmol .tmol params file (typing only — input ligand coordinates are scored as...
`ligand_params_files`	`Optional[list[str]]`	`None`	Pre-made tmol .tmol params file(s) for the ligand(s). Most deterministic typing route; bypasses SMILES generation.
`fix_backbone`	`bool`	`False`	Freeze backbone and minimise sidechains only (default: all free).

Metric Blocks¶

Blocks that compute metrics on existing structures — RMSD, contacts, energetics, clustering.

aromatic_neighbors¶

Count aromatic residues near ligand atoms.

Outputs and interpretation: Counts aromatic residues near selected ligand atoms under the configured distance cutoff. This is a geometric neighborhood count, not a pi-stacking energy or proof of a favorable interaction.

Class: AromaticNeighborsBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`distance_threshold`	`float`	`6`	Distance threshold in Angstroms for finding aromatic neighbors
`ligand_atom_names`	`Optional[list[str]]`	`None`	Optional canonical ligand atom names to use as the distance query. None counts neighbors around the complete ligand.

chirality_check¶

Per-row chirality check vs a reference ligand.

Outputs and interpretation: Compares assigned per-atom CIP stereochemical codes with a reference ligand. A matching code requires a resolvable atom correspondence; missing or unassigned centers should not be assumed correct.

Class: ChiralityCheckBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`reference_cif`	`str`	required	Path to reference ligand structure (CIF/PDB) with bond records. Atom names in this reference must match those in the ...
`ligand_chain`	`Optional[str]`	`None`	Chain ID for the ligand atoms in row structures. None auto-detects via the `hetero` annotation (any atom with hetero=...

clash¶

Count non-bonded backbone heavy-atom clashes in a structure.

Outputs and interpretation: Counts backbone-heavy-atom pairs (N/CA/C/O) closer than the cutoff in Å, excluding atoms in the same residue and consecutive residues of one chain. This is a geometric clash screen, not a full steric energy calculation.

Class: ClashMetricsBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`cutoff`	`float`	`1.5`	Distance (A) below which a non-bonded backbone atom pair counts as a clash.
`backbone_only`	`bool`	`True`	Restrict to backbone heavy atoms (N/CA/C/O). Set False to use all heavy atoms — note this is not comparable across ge...
`sequence_separation`	`int`	`1`	Residues this far apart or closer in the same chain are treated as bonded and excluded. 1 excludes intra-residue and ...
`exclude_hetero`	`bool`	`True`	Drop hetero (ligand/cofactor) atoms before scoring. Required for correctness whenever a ligand is a canonical amino a...

contact_ms¶

Compute contact molecular surface between binder and target chains.

Outputs and interpretation: Distance-weighted contact molecular surface in Å²: the target surface buried by the selected binder chain under the py_contact_ms algorithm. It measures interface contact extent, not binding energy.

Class: ContactMSBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`binder_chains`	`list[str]`	required	Chain IDs for the binder
`target_chains`	`list[str]`	required	Chain IDs for the target
`target_atom_names`	`Optional[list[str]]`	`None`	Restrict CMS aggregation to these target atom names (e.g. ['C12', 'O3']). The molecular surface is still generated fr...
`target_residues`	`Optional[list[str]]`	`None`	Restrict CMS aggregation to these target residues, as chain+number identifiers (e.g. ['B56', 'B121']). Use for epitop...

dssr_motif_base_pairing¶

DSSR pairing and native classification retention for a mapped motif.

Outputs and interpretation: Compares DSSR base-pair assignments and native pairing classifications for motif residues mapped by diffused_index_map. Per-residue pairing and aggregate retention diagnostics preserve the full native/predicted pair annotations; SNAP is not used.

Class: DSSRMotifBasePairingBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`x3dna_dssr_bin`	`Optional[str]`	`<computed at runtime>`	Path to x3dna-dssr. Defaults to $X3DNA_BIN_DIR/x3dna-dssr.
`x3dna_home`	`Optional[str]`	`<computed at runtime>`	Value exported as $X3DNA while invoking x3dna-dssr.
`reference`	`str`	required	Native RNA/DNA structure used to derive partner annotations.
`residues`	`list[str]`	required	Reference motif residue ids to report, e.g. ['A10', 'B4'].
`continue_on_error`	`bool`	`False`	Return per-row status/error diagnostics instead of aborting the metric block when predicted-structure DSSR processing...
`which`	`list[int]`	`[0, -1]`	Structures carrying diffused_index_map and predicted coordinates.

dssr_snap_metrics¶

Run DSSR + SNAP on each row's last structure and attach 11 metrics.

Outputs and interpretation: Counts protein–DNA hydrogen bonds classified as major-groove, minor-groove, or backbone contacts, plus hydrogen-bond support and SNAP interaction categories. SNAP-derived fields are missing if SNAP fails; missing values are not zero interactions. See the implementation for all 11 keys.

Class: DSSRSnapMetricsBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`x3dna_dssr_bin`	`Optional[str]`	`<computed at runtime>`	Path to the x3dna-dssr binary. Defaults to `$X3DNA_BIN_DIR/x3dna-dssr` from the environment (configure these in .en...
`x3dna_snap_bin`	`Optional[str]`	`<computed at runtime>`	Path to the x3dna-snap binary. Defaults to `$X3DNA_BIN_DIR/x3dna-snap` from the environment.
`x3dna_home`	`Optional[str]`	`<computed at runtime>`	Value to export as $X3DNA when invoking the binaries (the 3DNA install root containing config/, lib/, etc.). Defaults...
`skip_snap`	`bool`	`False`	If True, skip SNAP entirely (fast path) — SNAP-derived metrics will be NaN.

foldseek¶

Cluster structures by structural similarity using Foldseek.

Outputs and interpretation: Groups structures by structural similarity at the requested Foldseek thresholds. Cluster counts describe diversity under that clustering protocol, rather than a per-design correctness or activity score.

Class: FoldseekBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`foldseek_exec`	`str`	`<computed at runtime>`	Path to foldseek binary
`tm_thresholds`	`list[float]`	`[0.5, 0.6, 0.7, 0.8]`	TM-score thresholds for clustering. Results are output as cluster_id. and n_clusters. for each.
`alignment_type`	`int`	`1`	1=structure only, 2=structure+sequence
`min_seq_id`	`float`	`0.0`	Min sequence identity for clustering (alignment_type=2 only)
`multimer_tm_threshold`	`float`	`0.65`	TM threshold for multimer clustering
`interface_lddt_threshold`	`float`	`0.65`	Interface lDDT threshold for multimer clustering
`chains`	`Optional[list[str]]`	`None`	Chain IDs to cluster on. If None, all chains are used.
`group_by_task`	`bool`	`True`	If True, cluster within each task group separately (uses the first structure's metadata.task field).

hbond_metrics¶

Compute hydrogen bond metrics between ligand and protein.

Outputs and interpretation: HBPLUS identifies hydrogen bonds between ligand/motif and surrounding protein atoms. Outputs count the selected interactions, optionally by ligand atom group. Counts depend on the input protonation and HBPLUS geometry rules and are not hydrogen-bond energies.

Class: HBondBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`atom_groups`	`Optional[dict]`	`None`	Mapping of group name to list of ligand atom names. When provided, per-group h-bond counts are included in the output.
`verbose`	`bool`	`False`	When True, output also includes a `pairs` list with per-hbond atom-level detail (protein_residue, protein_atom, ligan...
`hbplus_path`	`Optional[str]`	`None`	Explicit HBPLUS executable. Resolved once during setup and propagated to workers so they do not depend on worker-loca...

ipsae¶

Compute ipSAE between a binder chain and one or more target chains.

Outputs and interpretation: Interface predicted Score from Aligned Errors summarizes inter-chain confidence from a predictor's PAE matrix using the configured distance/PAE cutoffs. Higher confidence does not prove binding. Missing PAE cannot be treated as a passing score. See the upstream implementation.

Class: IpSAEBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`binder_chain`	`str`	`'A'`	Chain ID of the designed binder. Default 'A' matches RFD3 output.
`target_chains`	`Optional[list[str]]`	`None`	Chain IDs of the target. If None, all non-binder chains in the prediction are used.
`pae_cutoff`	`float`	`10.0`	PAE threshold for inclusion in the score (Å). Default 10.0 per Park & Cong.

ligand_clash¶

Reject severe protein-ligand heavy-atom penetrations cheaply.

Outputs and interpretation: n_clashes counts selected protein–ligand heavy-atom pairs separated by less than cutoff Å; minimum_distance and closest_pair identify the nearest contact. This simple distance test does not distinguish intentional covalent bonds.

Class: LigandClashMetricsBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`ligand_resnames`	`list[str]`	required	Ligand residue names to score, e.g. ['ADN'].
`cutoff`	`float`	`2.0`	Distance (A) below which a severe penetration is counted.

ligand_ddg¶

Compute Rosetta DDG (binding energy) for protein-ligand complexes.

Outputs and interpretation: ddg_norepack is Rosetta's bound-minus-separated score across jump 1, in Rosetta energy units; more negative values favor the bound state under that score function. The bundled protocol uses beta_nov16, minimizes before scoring, and disables repacking in the Ddg filter (repack=0, one repeat). It is not an experimental binding free energy or kcal/mol estimate. Check that jump 1 separates the intended partners. See the exact protocol and flags; overriding either changes the interpretation.

Class: LigandDDGBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`rosetta_bin`	`str`	`<computed at runtime>`	Path to rosetta_scripts binary
`protocol`	`str`	`'<checkout>/src/blocks/scripts/ddg_protocol.xml'`	Path to Rosetta XML protocol file
`flags`	`str`	`'<checkout>/src/blocks/scripts/flags_rescore_genpot'`	Path to Rosetta flags file
`dalphaball_bin`	`Optional[str]`	`<computed at runtime>`	DAlphaBall executable required by the bundled Holes filter; also searches PATH.
`override_param_file`	`Optional[str]`	`None`	Override path to ligand .params file for -extra_res. Bypasses auto-resolution from ligand name.
`override_ligand_names`	`Optional[list[str]]`	`None`	Override ligand 3-letter codes (e.g. ['FAD', 'IAI']). Bypasses auto-detection from structure metadata. The first matc...
`ligand_map`	`Optional[dict]`	`None`	Rename ligand residues in the atom array before detection and PDB writing. Maps old res_name to new res_name, e.g. {'...

ligand_ddg_soft¶

Soft-repulsive ddG (pack hard, score soft, trimmed mean over replicates).

Outputs and interpretation: ddg_soft is the mean bound-minus-separated Rosetta score after packing with the hard beta_nov16 function and scoring with beta_nov16_soft. With at least three replicates the best and worst are dropped; ddg_soft_std reports the retained replicate spread. Values are Rosetta energy units, not measured binding free energies; do not compare numerically with the hard-score ligand_ddg protocol.

Class: LigandDDGSoftBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`replicates`	`int`	`5`	Pack+score cycles. ≥3 → trimmed mean (drop best+worst).
`binder_chain`	`str`	`'A'`	Chain ID of the binder.
`ligand_chain`	`str`	`'B'`	Chain ID of the ligand.
`interesting_atoms`	`Optional[list[str]]`	`None`	Ligand heavy-atom names to keep (plus their bonded H). Params is generated from the subset.
`ph`	`float`	`7.4`	Target pH for Dimorphite-DL protonation of H-less ligands.
`override_params_file`	`Optional[str]`	`None`	Pre-made .params file. Bypasses protonation and params generation — ligand H must already match.
`ref_smiles`	`Optional[str]`	`None`	Reference SMILES for rebuilding the ligand bond graph when atom names don't follow CCD convention (common in diffusio...
`cleanup_params_cache`	`bool`	`True`	Delete the shared params cache when the last chunk exits.

ligand_pocket¶

Quantify ligand CMS, enclosure, shell depth, and contact distribution.

Outputs and interpretation: Combines contact molecular surface, heavy-atom contact coverage, pocket shell/enclosure, and polar-atom neighborhood counts for the selected ligand. Surface uses Å² and distance cutoffs use Å; contact counts and coverage summarize geometry, not interaction energy.

Class: LigandPocketMetricsBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`ligand_resnames`	`list[str]`	required	Ligand residue names to measure, e.g. ['ADN'].
`core_atom_names`	`list[str]`	`<computed at runtime>`	Ligand core atoms that must all receive protein contact.
`tail_polar_atom_names`	`list[str]`	`<computed at runtime>`	Ligand tail atoms requiring explicit polar support.
`contact_distance`	`float`	`5.0`	Protein-heavy-atom distance defining coverage and shell.
`polar_distance`	`float`	`3.6`	Maximum protein N/O/S distance defining polar support.
`clash_distance`	`float`	`2.0`	Protein-ligand heavy-atom distance below which a severe clash is counted.

metal3d¶

Score each row's structure for transition-metal binding sites.

Outputs and interpretation: A learned three-dimensional probability field for transition-metal binding sites. Predicted site confidence and position describe the model's output, not experimentally confirmed metal occupancy.

Class: Metal3DBlock · Input: True · Resources: partition=cpu, mem=16GB, time=5:00:00 · Container: metal3d

Parameter	Type	Default	Description
`metalbinding`	`bool`	`True`	If True, only voxelize residues whose sidechains can coordinate metals (HIS/CYS/ASP/GLU/ASN/GLN/MET). If False, scan ...
`pthreshold`	`float`	`0.1`	Minimum probability for a voxel to seed a clustered site. Lower values produce more (noisier) sites.
Additional parameters
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`cluster_threshold`	`float`	`7.0`	Maximum distance (Å) between probe voxels in the same cluster. Larger values merge nearby sites into one.
`batch_size`	`int`	`128`	Inference batch size (number of voxel boxes).
`write_cube`	`bool`	`False`	If True, also write the volumetric probability map as a .cube file per input. Off by default — files can be large.
`compile_as_metric`	`bool`	`True`	If True (default), attach Metal3D metrics to the parent row without adding the probefile to the lineage — same conven...

motif_confidence¶

Motif vs scaffold pLDDT and PAE of an oracle refold.

Outputs and interpretation: Summarizes predictor pLDDT and PAE separately for the mapped motif and scaffold. Higher pLDDT and lower PAE indicate model confidence, not experimentally validated accuracy; the upstream confidence scale and residue mapping must be preserved.

Class: MotifConfidenceBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`which`	`list[int]`	`[0, -1]`	Two indices into each row's structures list: the structure carrying diffused_index_map, and the prediction to score.
`residues`	`Optional[list[str]]`	`None`	Reference residue ids to treat as the motif, e.g. ['A10', 'A11']. Defaults to every polymer entry in diffused_index_map.
`chain`	`Optional[str]`	`None`	Designed chain to score. Defaults to the chain the mapped motif residues land in.

motif_geometry¶

Named distances and angles between motif atoms, addressed by reference id.

Outputs and interpretation: Reports named distances in Å and angles between selected motif atoms addressed using reference residue identifiers. Compare each quantity against the intended geometry rather than treating all distances as smaller-is-better.

Class: MotifGeometryBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`measurements`	`Dict`	required	Name -> atom specs as '.', e.g. {'contact': ["A24.N1", "B4.O2'"]}. Two specs give a ...
`which`	`list[int]`	`[0, -1]`	Two indices into each row's structures list: the structure carrying diffused_index_map, and the structure to measure.

mpnn_score¶

Attach MPNN per-sequence confidence (from FASTA) as metrics to each row.

Outputs and interpretation: Copies per-sequence overall_confidence, seq_rec, and optional ligand_confidence from MPNN outputs. Confidence is an upstream model ranking signal and sequence recovery measures agreement with the input sequence; neither establishes folding or binding success. Backend definitions can differ.

Class: MpnnScoreBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...

peppr¶

Compute structural evaluation metrics using peppr.

Outputs and interpretation: PEPPR computes the requested structure and interface comparison metrics against a reference; there is no single universal PEPPR score. Interpret each returned metric using its upstream definition and retain the atom/chain mapping and reference used for the calculation.

Class: PepprBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`metrics`	`list[PepprMetric]`	`[<PepprMetric.RMSD: 'rmsd'>, <PepprMetric.TM_SCORE: 'tm_score'>, <PepprMetric.LDDT: 'lddt'>, <PepprMetric.LDDT_PLI: 'lddt_pli'>, <PepprMetric.FNAT: 'fnat'>, <PepprMetric.IRMSD: 'irmsd'>, <PepprMetric.LRMSD: 'lrmsd'>, <PepprMetric.DOCKQ: 'dockq'>]`	Which metrics to compute. Options: rmsd, tm_score, lddt, lddt_pli, fnat, irmsd, lrmsd, dockq
`rmsd_threshold`	`float`	`2.0`	RMSD threshold (Angstroms) for MonomerRMSD classification
`tolerate_exceptions`	`bool`	`True`	If True, continue when individual metrics fail (e.g., lDDT-PLI with no ligand). Failed metrics are stored as None.

plip¶

Profile protein-ligand interactions using PLIP.

Outputs and interpretation: PLIP classifies protein–ligand contacts into interaction types such as hydrogen bonds, hydrophobic contacts, and salt bridges. Returned counts and optional details depend on PLIP's geometric and chemical rules; they are not binding energies. See PLIP.

Class: PLIPBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`ligand_ids`	`Optional[list[str]]`	`None`	Filter to specific binding sites by PLIP ID (format: 'HetID:Chain:Position', e.g. 'LIG:A:301'). When None, all detect...
`per_residue`	`bool`	`False`	Include per-residue total interaction counts (e.g. per_res.A42 = 3).
`include_details`	`bool`	`False`	Include per-ligand interaction breakdown when multiple binding sites are present.
`verbose`	`bool`	`False`	When True, output also includes `per_interaction_details`: for each interaction type, a list of dicts with atom-level...

pocket_coulomb¶

Electrostatic complementarity between charged ligand atoms and pocket residues.

Outputs and interpretation: An approximate charge-based pocket interaction score using the configured charge and distance treatment. Its scale depends on those settings and is not an experimental affinity.

Class: PocketCoulombBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`pocket_radius`	`float`	`8.0`	CB distance cutoff (A) for defining pocket residues around the ligand
`distance_floor`	`float`	`1.5`	Minimum distance (A) to prevent singularities from clashes
`distance_cutoff`	`float`	`12.0`	Ignore charged protein atoms beyond this distance (A) from any charged ligand atom
`power`	`float`	`2.0`	Distance exponent for coulomb-like decay (2.0 = implicit solvent screening)
`ref_smiles`	`Optional[str]`	`None`	Reference SMILES for rebuilding ligand bond graph when atom names don't follow CCD convention (diffusion model outputs)
`per_residue`	`bool`	`False`	Include per-residue score breakdown in output metrics

ppi_ddg¶

Protein-protein binding energy via FastRelax + ddG.

Outputs and interpretation: Protein–protein binding score computed with FastRelax and a bound-versus-separated ddG calculation. These are Rosetta energy units, not experimental affinities. Partner selections, relaxation, and score-function settings must match for comparisons; consult the implementation for the returned energy terms.

Class: PPIDDGBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`binder_chain`	`str`	`'A'`	Chain ID of the binder (translated far away for unbound score).
`target_chain`	`str`	`'B'`	Chain ID of the target (stays in place).
`relax_rounds`	`int`	`1`	FastRelax standard_repeats. 1 is fast (~1–3 min/structure on CPU); 3–5 gives lower, more converged energies.
`replicates`	`int`	`1`	Independent relax + ddG cycles. Mean and std are reported. Each additional replicate adds one full relax run.
`constrain_to_start_coords`	`bool`	`True`	Harmonic coordinate constraints during FastRelax — keeps the backbone close to the AF3-predicted geometry.
`constraint_weight`	`float`	`1.0`	Coordinate constraint weight. Higher = stays closer to input coords; lower = more relaxation freedom.

rasa¶

Compute solvent accessible surface area (SASA/RASA) for ligand atoms.

Outputs and interpretation: Per-atom relative solvent accessibility is computed as solvent-accessible area divided by an element-specific reference area. ligand_rasa is the sum of these fractions, not a fraction bounded by one; mean_ligand_atom_rasa is their mean. Smaller values indicate burial. Keep probe radius, atom groups, and backbone-only settings fixed when comparing designs.

Class: RASABlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`atom_groups`	`Optional[dict]`	`None`	Mapping of group name to list of ligand atom names. When provided, per-group RASA aggregates are included in the output.
`probe_radius`	`float`	`1.4`	Radius of the solvent probe in Angstroms.
`backbone_only`	`bool`	`False`	If True, strip protein sidechains before SASA so ligand accessibility is measured against backbone atoms (N, CA, C, O...

reference_motif_rmsd¶

RMSD of the designed motif against the structure it was taken from.

Outputs and interpretation: Motif atom RMSD in Å against the original input reference, with correspondence resolved through diffused_index_map. Smaller values indicate closer motif geometry; mapping coverage and atom selection matter.

Class: ReferenceMotifRMSDBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`reference`	`str`	required	Path to the canonical structure the motif was taken from.
`which`	`list[int]`	`[0, -1]`	Two indices into each row's structures list: the structure carrying diffused_index_map, and the structure to score. D...
`residues`	`Optional[list[str]]`	`None`	Reference residue ids to score, e.g. ['A10-12', 'B4']. Defaults to every entry in diffused_index_map.
`ligands`	`list[str]`	`<computed at runtime>`	Ligand/ion residue names to include, e.g. ['MN']. RFD3NA leaves non-polymer components out of diffused_index_map, so ...

rmsd¶

Compute alignment and RMSD metrics between two designs in the lineage.

Outputs and interpretation: RMSD is the root mean square atom displacement in Å after fitting the selected alignment atoms. Output names such as bkbn_align.motif_allatom_rmsd identify the fit selection and the measured selection. Lower means closer geometry; compare only matched atoms and selections.

Class: RMSDBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`which`	`list[int]`	`[0, -1]`	Indices into each row's structures list selecting the two structures to compare. Default [0, -1] compares first vs la...
`alignment_methods`	`list[SELECTIONS]`	`[<SELECTIONS.BKBN: 'bkbn'>, <SELECTIONS.MOTIF_BKBN: 'motif_bkbn'>]`	List of selections to use for computing alignment transformations
`rmsd_targets`	`list[SELECTIONS]`	`[<SELECTIONS.BKBN: 'bkbn'>, <SELECTIONS.MOTIF_ALLATOM: 'motif_allatom'>, <SELECTIONS.LIGANDS: 'ligand'>]`	List of selections to compute RMSD on after alignment
`resolve_symmetries`	`bool`	`True`	Whether to resolve automorphisms for all-atom RMSD
`select_mask_by_name_pairing`	`list[SELECTIONS]`	`[<SELECTIONS.LIGANDS: 'ligand'>]`	List of RMSD targets where atom masks are paired by finding common atoms (via chain_id, res_id, atom_name) instead of...
`component_remap`	`Optional[dict]`	`None`	Remap component names for ligand lookup (e.g., {'LIG_B': 'B;N'}). Needed when upstream tools rename ligands and RMSD ...
`atom_groups`	`Optional[dict]`	`None`	Mapping of group name to list of ligand atom names. When provided, per-group ligand RMSD values are included in the o...

rna_connectivity¶

Real O3'-P backbone connectivity of a designed nucleic-acid chain.

Outputs and interpretation: Detects nucleic-acid backbone breaks from the O3′ of residue i to the P atom of residue i+1 using the configured distance criterion. Missing atoms and residue mapping affect which links can be evaluated.

Class: RNAConnectivityBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`chain`	`Optional[str]`	`None`	Chain to walk. Defaults to every chain carrying C3' atoms.
`max_bond_distance`	`float`	`2.0`	O3'(i)-P(i+1) distance above which the junction is a break. The bond is ~1.6 A.
`which`	`str \\| list[int]`	`'last'`	Which structure(s) of each row to measure. Measure the diffusion output, not a refold: an oracle rebuilds a continuou...

rna_ss_f1_mcc¶

Motif-restricted secondary-structure F1 / MCC against reference pairs.

Outputs and interpretation: F1 and Matthews correlation coefficient compare predicted/reference RNA secondary-structure pair assignments. The assigned base pairs and matching scheme define positives and negatives; inspect unresolved residues before comparing results.

Class: RNASecondaryStructureF1Block · Input: True · Resources: partition=cpu, mem=4GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`predictions_csv`	`str`	required	RNet2 runner output CSV. Must carry 'predicted_dbn' and the key column named by 'join_on'.
`join_on`	`str`	`'design_path'`	Column shared with the handler used to match a prediction to its design. Row position is NOT used - the runner preser...
`reference_pairs`	`list[list[str]]`	`<computed at runtime>`	Canonical (WC + G.U) reference pairs as reference residue ids, e.g. [["A5", "A10"], ["A12", "A25"]]. Resolved through...
`pseudoknot_pairs`	`list[list[str]]`	`<computed at runtime>`	Reference pairs excluded from R because they are pseudoknotted.
`compare_to_vienna`	`bool`	`True`	Also report whole-sequence F1 between RNet and ViennaRNA's MFE on the same design. Cross-method 2D disagreement is a ...

rna_thermo¶

ViennaRNA Boltzmann scoring of a required base-pair set.

Outputs and interpretation: ViennaRNA estimates p_motif = exp(-(G_constrained - G_unconstrained) / RT), the ensemble probability of the required motif base pairs under its thermodynamic model. Pseudoknots are excluded and reported with partial status; the model lacks an Mg²⁺ term. Use it to compare designs within a target, not as an absolute measure of full-complex folding.

Class: RNAThermoBlock · Input: True · Resources: partition=cpu, mem=4GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`required_pairs`	`list[list[str]]`	`<computed at runtime>`	Base pairs the motif requires, as reference residue ids resolved through diffused_index_map, e.g. [["A10", "A24"], ["...
`pseudoknot_pairs`	`list[list[str]]`	`<computed at runtime>`	Pairs that form pseudoknots. Excluded from the constraint because ViennaRNA's model cannot represent them; reported a...
`chain`	`Optional[str]`	`None`	Restrict to one chain id. None uses every RNA residue.

rotamer_metrics¶

Compute rotamer quality metrics for protein structures.

Outputs and interpretation: Reports sidechain rotamer quality against the selected library/model. Percentile and probability fields are different quantities; compare using the same backend and residue selection.

Class: RotamerMetricsBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`backend`	`Literal`	`'dunbrack'`	Scoring backend: 'peppr' (Top8000 library via peppr) or 'dunbrack' (backbone-dependent Dunbrack library).
`motif_only`	`bool`	`True`	Only score residues annotated as motif (is_motif_atom). If the structure has no motif annotation, all residues are sc...
`include_rama`	`bool`	`True`	Also compute Ramachandran scores (peppr backend only).
`detailed_output`	`bool`	`False`	Include per-residue-type breakdown of scores.
`per_residue_output`	`bool`	`True`	Include per-original-residue scores using diffused_index_map metadata to map back to original residue identifiers (e....

secondary_structure¶

Per-fold SS composition + loop/pLDDT disorder, globally and around the ligand.

Outputs and interpretation: Reports secondary-structure composition and loop/confidence disorder summaries globally or around a ligand pocket. Confidence interpretation requires the input B-factor field to carry the intended pLDDT values.

Class: SecondaryStructureBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`pocket_cutoff`	`float`	`8.0`	Å cutoff for a residue to be considered in the pocket (any non-H protein atom within this distance of any ligand atom).
`plddt_threshold`	`float`	`70.0`	Per-residue mean b_factor below this counts as low-confidence (pLDDT semantics; b_factor in AF3/Boltz/RF3 CIFs carrie...
`plddt_min_signal`	`float`	`50.0`	Auto-scaling rules for b_factor → pLDDT (0-100): - max(bf) ≤ 1.5 → assume 0-1 scale (RF3); multiply by 100....
`verbose`	`bool`	`False`	When True, also emit `ss_string`, segment topology, and a `residue_records` list with per-residue {residue_id, resnam...
`chains`	`Optional[list[str]]`	`None`	Restrict SS computation to these chain IDs (e.g. ['B']). When None, all chains are included.

sequence_identity_gate¶

Hard novelty gate on scaffold-region global sequence identity.

Outputs and interpretation: Checks identity against a reference sequence collection and reports the nearest match/gate verdict. Identity thresholds describe sequence similarity, not structural or functional novelty by themselves.

Class: SequenceIdentityGateBlock · Input: True · Resources: partition=cpu, mem=4GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`references`	`str`	required	FASTA of natural sequences to compare against - the native construct plus Rfam/RNAcentral homologs.
`threshold`	`float`	`0.4`	Maximum scaffold identity a design may have and still pass. Set per campaign; 0.40 is the RNA motif-scaffolding bench...
`exclude_motif`	`bool`	`True`	Score the scaffold region only. Leave True - motif identity is 100% by construction and including it makes the gate u...
`chain`	`Optional[str]`	`None`	Restrict to one chain id. None uses every RNA residue.
`compute_intra_identity`	`bool`	`False`	Also report the maximum identity to any OTHER design in this chunk, which measures whether the designer explored or e...

symmetry¶

Measure how closely an assembly obeys its declared cyclic point group.

Outputs and interpretation: Measures deviation of a homo-oligomeric assembly from cyclic point-group symmetry. Atom selection and subunit correspondence define the geometric comparison; symmetry alone does not establish stability.

Class: SymmetryMetricsBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`point_group`	`str`	`'C3'`	Expected cyclic point group, e.g. 'C3'. Only Cn is supported; dihedral and polyhedral groups need an explicit subunit...
`chains`	`Optional[list[str]]`	`None`	Protomer chain IDs. Defaults to every protein chain in the structure, alphabetically. Name them explicitly to score o...
`backbone_only`	`bool`	`True`	Superimpose on backbone N/CA/C/O only. Set False to include side chains — only comparable when every protomer is full...
`exclude_hetero`	`bool`	`True`	Drop hetero (ligand/cofactor) atoms before scoring, so a bound cofactor never enters the protomer superposition.

tm_score¶

TM-score against an external reference structure.

Outputs and interpretation: TM-score compares a structure with an external reference fold using a length-normalized structural similarity score. Higher values indicate greater similarity; reference choice, aligned chains, and normalization determine comparability.

Class: TMScoreBlock · Input: True · Resources: partition=cpu, mem=4GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`reference`	`str`	required	Path to the reference structure (PDB/CIF).
`reference_chain`	`Optional[str]`	`None`	Restrict the reference to one chain id.
`chain`	`Optional[str]`	`None`	Restrict the target to one chain id.

tm_score_rna¶

TM-score_RNA against a reference fold, computed by US-align.

Outputs and interpretation: RNA structural similarity against a reference, using the block's selected nucleotide anchor atoms and alignment/normalization settings. Higher TM-score means closer overall fold; it does not certify base-pair or catalytic geometry.

Class: TMScoreRNABlock · Input: True · Resources: partition=cpu, mem=4GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`reference`	`str`	required	Path to the reference structure (the native fold).
`reference_chain`	`Optional[str]`	`None`	Restrict the reference to one chain. None keeps every chain and aligns them as one unit (-ter 0).
`chain`	`Optional[str]`	`None`	Restrict the design to one chain. None uses all.
`expected_reference_length`	`Optional[int]`	`None`	Nucleotide count the reference should parse to. Checked at setup - a mis-parsed reference silently changes d0 and eve...
`fixed_correspondence`	`bool`	`False`	Score the residue-index correspondence as given (-TMscore 1) instead of searching for an alignment. Only meaningful w...
`mol`	`str`	`'RNA'`	US-align molecule type: RNA or prot.
`usalign_binary`	`Optional[str]`	`None`	US-align executable name or path. Defaults to the USALIGN_BIN setting, the venv's bin, then PATH.
`min_reference_length`	`int`	`30`	Refuse a reference shorter than this. d0 goes to zero near L=18, so TM-score_RNA is undefined on short selections - u...

Utility Blocks¶

Blocks for data manipulation — ingestion, filtering, format conversion, annotation.

annotate¶

Annotate molecular structures with interface, hbond, and ligand info.

Class: AnnotateBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`annotations`	`list[Literal]`	required	Annotations to compute; select explicitly to avoid unnecessary work.
`overwrite`	`bool`	`True`	Whether to overwrite existing annotations
`interface_config`	`InterfaceConfig`	`<computed at runtime>`	Configuration for interface residue annotation

filter¶

Filter pipeline rows by metric range filters or a predicate function.

Class: FilterBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`expr`	`Optional[str]`	`None`	Boolean filter expression with and/or/not and parentheses over metric keys, e.g. 'af3.min_pae_interaction <= 10 and (...
`filters`	`Optional[dict]`	`None`	Metric key -> MetricFilter(min/max). Uses Polars-native fast path.
`filter_groups`	`Optional[dict]`	`None`	Named groups of filters combined with OR logic. Filters within each group use 'filter_groups_inner_operator'; groups ...
`filter_groups_inner_operator`	`Literal`	`'and'`	How filters within each filter_group are combined before groups are OR'd together.
`function`	`Optional[Callable]`	`None`	Predicate function: metrics dict -> bool (True = keep)
`check_column_exists_only`	`bool`	`False`	When True, validate that filter columns exist and contain valid floats (not NaN) but do not prune rows. Useful for te...

ingest¶

Ingest data from various sources (dir, csv, dataframe, sequences, or compile).

Class: IngestBlock · Input: False · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`source`	`str \\| PipelineDataHandler`	required	Data source to ingest. When compile_as is provided, this should be a comma-separated list of directory paths. Otherwi...
`compile_as`	`Optional[str]`	`None`	If provided, treat source as comma-separated directories and compile using the specified block (e.g., 'rfd3', 'mpnn')

materialize¶

Write stored structures to disk.

Class: MaterializeBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`output_dir`	`Optional[str]`	`None`	Target directory for materialized files. Defaults to sandbox_dir/materialized.
`nest_by_step`	`bool`	`True`	Organize files into subdirectories by step name (e.g. output_dir/rfd3/, output_dir/af3/).

mutate_residues¶

Expand each input structure into named residue-mutation variants.

Class: MutateResiduesBlock · Input: True · Resources: partition=cpu, mem=4GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`variants`	`dict`	required	Named mutation sets, e.g. {'a73f': {'A73': 'PHE'}, 'double': {'A38': 'PHE', 'A73': 'PHE'}}.

nanoparticle_prepare¶

Align antigen clash reference and preserve a C3 scaffold's symmetry axis

Requires RFdiffusion motif correspondence. Emits an asymmetric unit with motif metadata and preserved radial placement; rows exceeding the clash limit are removed. Invalid symmetry or missing correspondence fails explicitly.

Class: NanoparticlePrepareBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`motif_pdb`	`str`	required	The original RFdiffusion C3 antigen stub
`clash_ref_pdb`	`str`	required	Full antigen stalk in the stub's coordinate frame
`max_clashes`	`int`	`8`	—
`max_c3_rmsd`	`float`	`2.0`	—

order_collection¶

Extract sequences, check ordered status, and collect prediction metrics.

Class: OrderCollectionBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`ordered_sequences`	`list[str]`	`<computed at runtime>`	Ordered sequence fingerprints (from final filter)
`prediction_step`	`str`	`''`	Prediction step name to extract metrics from

package_submission¶

Write a validated designs.csv (and optional run_meta.json).

Class: PackageSubmissionBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`output_dir`	`str`	required	Directory to write designs.csv (and run_meta.json) into
`expected_count`	`Optional[int]`	`None`	Exact number of designs required; None accepts any count
`sequence_key`	`str`	`'sequence'`	Column holding the amino-acid sequence. The default resolves standard sequence_ columns and structure-backed p...
`allow_duplicates`	`bool`	`False`	Permit repeated sequences (off: duplicates are a violation)
`metadata`	`Optional[dict]`	`None`	Written verbatim to run_meta.json; omitted when None
`extra_columns`	`Optional[list[str]]`	`None`	Additional columns to carry into designs.csv alongside sequence

remap_ligand¶

Rename predicted ligand atom/residue names to match a reference CIF.

Class: RemapLigandBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`reference_cif`	`str`	required	Path to a ligand-only CIF with canonical atom names and _chem_comp_bond records.
`target_name`	`Optional[str]`	`None`	Desired residue name in outputs (e.g. 'B;N'). Read from the reference CIF if None.
`atom_name_mapping`	`Optional[dict]`	`None`	Optional explicit predicted-to-reference atom-name mapping. When provided, bypass graph discovery and validate each r...
`core_atoms`	`Optional[list[str]]`	`None`	Reference atom names of the rigid core. When given, the RMSD used for the substructure-match tiebreak is computed onl...
`buried_atoms`	`Optional[list[str]]`	`None`	Backward-compatible alias for core_atoms.
`sample_size`	`int`	`64`	Scan this many structures first before streaming the rest of the chunk.
`rmsd_ok`	`float`	`2.0`	Stop on the first mapping with RMSD below this threshold (Å).
`num_workers`	`Optional[int]`	`0`	Parallel workers for the atom-mapping scan. 0 (default) runs serial — avoids a fork() deadlock when the scan runs ins...
`emit_metric`	`bool`	`True`	Store the match RMSD under '.match_rmsd' on every remapped row.

restore_sidechains¶

Restore explicitly selected side-chain conformers from an earlier step.

Class: RestoreSidechainsBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`positions_json`	`str`	required	JSON mapping reference structure paths to residue positions.
`reference_index`	`int`	`0`	Index of the reference structure in each lineage row.
`align_backbone`	`bool`	`False`	Rigidly align each reference residue's backbone onto the current backbone before copying its side chain. Use for pred...

rotamer_relax¶

PyRosetta rotamer relaxation for sidechain optimization.

Class: PyrosettaRotamerRelaxBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...

rotamer_sweep¶

Mutate named residues and enumerate explicit chi-angle combinations.

Class: SidechainRotamerSweepBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`mutations`	`dict`	required	Residue identities to install, e.g. {'A12': 'PHE'}.
`rotamers`	`dict`	required	Per-position chi vectors in degrees. Every Cartesian combination is emitted as a separate structure.
`max_combinations`	`int`	`1000`	Safety cap on emitted combinations per input structure.

select_chains¶

Keep specified chains, preserving residue IDs and parent lineage

Missing chains raise an error. The output is a new CIF; motif residues, atom selections and diffusion correspondence are restricted to retained chains. This supports a true protein-only prediction branch when a complex also contains a protein or peptide partner, which ligand removal alone would retain.

Class: SelectChainsBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`chains`	`list[str]`	required	—

select_diverse¶

Keep a subset of rows no two of which exceed the similarity cutoffs.

Class: SelectDiverseBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`max_identity`	`float`	`0.5`	Reject a design sharing more than this identity with a kept one
`target_count`	`Optional[int]`	`None`	Stop once this many are selected; None keeps every non-conflicting design
`rank_by`	`Optional[str]`	`None`	Metric column to prefer when breaking conflicts (best first)
`rank_ascending`	`bool`	`True`	True when a lower rank_by value is better (e.g. PAE)
`sequence_key`	`str`	`'sequence'`	Column holding the amino-acid sequence. The default resolves standard sequence_ columns and structure-backed p...
`require_target`	`bool`	`False`	Raise if target_count cannot be met, rather than warning
`required_structure_steps`	`list[str]`	`<computed at runtime>`	Only select rows containing outputs from every named structure step (for example ['rf3']); use this to prevent unscre...

select_motif¶

Apply a user-defined function across structures to update motif annotations.

Class: SelectMotifBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`selector`	`Optional[Callable]`	`None`	Function mapping LoadedStructure -> list of residue IDs to fix
`metadata_key`	`str`	`'motif'`	Metadata key to write selected residues into
`append`	`bool`	`True`	Append to existing metadata key (True) or overwrite (False)

selective_sidechain_repack¶

Resolve local protein clashes without moving the backbone or ligand.

Class: SelectiveSidechainRepackBlock · Input: True · Resources: partition=cpu, mem=8GB, time=5:00:00

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`positions`	`list[str]`	`<computed at runtime>`	Protein residue IDs to repack, such as ['A42', 'A:73'].
`clash_distance`	`Optional[float]`	`None`	When set, also detect and repack each structure's residues in nonbonded protein-heavy-atom contacts below this distance.
`sequence_separation`	`int`	`1`	Same-chain residue separation excluded from clash detection.
`scorefxn_name`	`str`	`'beta_nov16'`	Rosetta score function used to select fixed-identity rotamers.
`extra_rotamers`	`bool`	`True`	Include extra chi1-chi4 sampling at the selected positions.

to_cif¶

Parse input structures and re-save as clean CIF files.

Class: ParseToCifBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...

to_pdb¶

Parse input structures and re-save as clean PDB files.

Class: ParseToPdbBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...

transform¶

Apply an atomworks Transform to each row of the pipeline.

Class: TransformBlock · Input: True · Resources: Not specified

Parameter	Type	Default	Description
`store_cleanup`	`bool`	`False`	When `store='catcif'`, delete sandbox output files after they have been consolidated into the archive. Ignored othe...
`transform`	`Any`	`None`	An atomworks Transform instance to apply
`extract_metrics`	`Optional[list[str]]`	`None`	Keys to extract from the transform output as metrics
`extract_metadata`	`Optional[list[str]]`	`None`	Keys to extract from the transform output as metadata updates
`write_structure`	`bool`	`False`	If True, write result['atom_array'] back as a new CIF file

Standard Parameters¶

The following parameters are shared across most blocks and are omitted from the per-block tables above for brevity.

Parameter	Description
`store`	Store structure file contents in a Delta Lake table for portable/archival runs. Default: `False`. Available on all blocks.
`command_template`	Override the container/binary command template. Pass a list of strings to run against a local install instead of the default container. Available on compiled blocks.
`cloud`	Set `True` to request Modal execution for a compiled block. Default: `False`; the engine backend can also select Modal. See cloud setup.