DeepRank-Ab
v1.0.2Score antibody and nanobody complexes with predicted DockQ.
DeepRank-Ab 1.0.2 scores antibody-antigen and nanobody-antigen complex models without requiring a reference structure. It returns predicted DockQ scores, ranked models, and flags for heavy-light chain contacts and potential steric clashes.
Runs cost 27 credits per minute of measured runtime. The minimum reservation is 27 credits, not a minimum final charge. Completed runs are charged in proportion to elapsed time, rounded up to a whole credit; unused reserved credits are returned. Set a spending limit before submission. Each batch job is metered separately.
| Input | Accepted values | Details |
|---|---|---|
| Complex structure | .pdb, .ent, .cif, .mmcif | A complex containing the antibody or nanobody and its antigen; maximum 50 MiB per file. |
| Ensembles | Multiple models in one structure file | Each model is scored within the same job. |
| Batch | Up to 10 structure files | Each file runs as a separate job. Uploads, saved files, ZIP batches and RCSB PDB-code fetching use the shared input controls. |
CIF/mmCIF files must be convertible to PDB while preserving model, chain, residue and atom identities, alternate conformations, coordinates, occupancy and temperature factors. Files that cannot preserve these fields are rejected. The original CIF and converted PDB remain separate downloads.
Filenames should contain letters, such as complex_001.pdb; purely numeric names such as 001.pdb can fail during result processing. A scoring run has a 25-minute execution limit.
| Parameter | Type | Default | Description |
|---|---|---|---|
Complex type (complex_type) | enum | auto | auto uses automatic chain detection; antibody requires heavy, light and antigen chain IDs; nanobody specifies that no light chain is present. |
Heavy / binder chain (heavy_chain_id) | string | Auto-detect (blank) | Optional chain ID override; required in antibody mode. Automatic detection uses ANARCI. |
Light chain (light_chain_id) | string | Auto-detect (blank) | Optional chain ID override; required in antibody mode and hidden in nanobody mode. |
Antigen chain (antigen_chain_id) | string | Auto-detect (blank) | Optional single chain ID override; required in antibody mode. Automatic detection includes and merges multiple antigen chains. |
Auto-detect supports partial overrides. When both heavy and antigen chains are specified, leaving the light chain blank means no light chain, rather than automatic light-chain detection. Antibody mode requires all three chain IDs. Nanobody mode always uses no light chain.
The Predictions table lists models in descending predicted DockQ order and supports copying and CSV download. Downloads also retain the generated result files:
| Download | Contents |
|---|---|
*_predictions.csv | Model identifiers, predicted DockQ scores and both quality flags. |
*_predictions.hdf5 | Prediction data and associated metadata. |
| Structure files | Original structures, split ensemble models and prepared PDB structures. |
| FASTA and JSON files | Sequences and antibody-region annotations. |
Graph HDF5 and .pt files | Molecular graphs and ESM sequence embeddings used for scoring. |
deeprank_ab.log | Calculation progress, warnings and diagnostic messages; captured diagnostics are also retained for execution failures and timeouts. |
provenance.json | Run settings, software and model identity, and input checksums. |
| Field | Meaning |
|---|---|
Model (pdb_id) | Identifier of the scored model. |
Predicted DockQ (predicted_dockq) | Predicted complex quality; higher values suggest a better modeled interface. It is a prediction, not DockQ calculated against a reference structure or a binding-affinity measurement. |
H-L contact flag (HL_contact_flag) | low_HL_contacts indicates few contacts between the heavy and light chains; ok passes this check; not_applicable means the check does not apply without both chains. |
VdW clash flag (vdw_clash_flag) | potential_clash indicates unusual van der Waals interaction energies; ok passes this check. |
The flags complement the score and do not establish that a model is experimentally correct. Software details are available in the DeepRank-Ab source documentation.
Characterize nanobody developability with sequence, structure, and surface-property metrics.
Predict non-germline residues and generate embeddings for paired or unpaired antibody sequences.
Analyze aggregation-prone regions in a protein structure.
Antibody language model for scoring, classification, embeddings and masked-residue filling
Predict pKa values of ionizable groups in proteins based on 3D structure.
Restore residues and generate antibody-specific representations or likelihood scores.
Model accuracy estimation and sequence ranking with AlphaFold
Predict MMP cleavage z-scores, evaluate substrates, and generate conditional peptides.
Automated atomic model building for cryo-EM maps
Sharpen and denoise cryo-EM maps.