How dauparas/proteinmpnn works
ProteinMPNN designs protein sequences from backbone structures and also scores sequences or reports residue probabilities. The principal workflow is structure parsing, optional chain/position and sequence-bias configuration, model inference using selected weights, then sequence or score/probability output. The README documents these capabilities; sampled source confirms the runner, parser, model components, and several constraint helpers. Some runner internals and optional modes were not sampled, so uncertain implementation links are omitted.
Main entry points: protein_mpnn_run.py Inference runner · training.py Training loop · parse_multiple_chains.py PDB parser · assign_fixed_chains.py Chain assignment
Components 18
Structure preparation
- PDB parser
helper_scripts/parse_multiple_chains.py - Parsed structure
Design constraints
- Chain assignment
helper_scripts/assign_fixed_chains.py - Fixed positions
helper_scripts/make_fixed_positions_dict.py - Tied positions
helper_scripts/make_tied_positions_dict.py - AA bias
helper_scripts/make_bias_AA.py - PSSM input
helper_scripts/make_pssm_input_dict.py
Inference and outputs
- Inference runner
protein_mpnn_run.py - Parsing and data utilities
protein_mpnn_utils.py - ProteinMPNN model
protein_mpnn_utils.py - Model weights
vanilla_model_weights - Sequences and scores
outputs
Model training
- Training loop
training/training.py - Training data utilities
training/utils.py - Training model
training/model_utils.py - Trained weights
training/exp_020/model_weights
Other components
- Designer
- PDB backbone
Connections 22
- Designer → PDB backbone: provides structure
- PDB backbone → PDB parser: parsed by
- PDB parser → Parsed structure: writes JSONL
- Parsed structure → Chain assignment: configures chains
- Parsed structure → Fixed positions: configures positions
- Parsed structure → Tied positions: configures ties
- Parsed structure → AA bias: configures bias
- Parsed structure → PSSM input: configures PSSM
- Designer → Inference runner: runs design or scoring
- Inference runner → Parsing and data utilities: uses parsing utilities
- Inference runner → ProteinMPNN model: runs inference
- Inference runner → Model weights: loads weights
- Inference runner → Parsed structure: reads structure data
- Inference runner → Chain assignment: reads chain settings
- Inference runner → Fixed positions: reads fixed positions
- Inference runner → Tied positions: reads tied positions
- Inference runner → AA bias: reads amino-acid bias
- Inference runner → PSSM input: reads PSSM settings
- Inference runner → Sequences and scores: writes sequences and scores
- Training loop → Training data utilities: loads training data
- Training loop → Training model: trains model
- Training loop → Trained weights: saves checkpoints