How christian-creator/protonpottsmpnn works
This repository supports pH-sensitive binder design with Proton-PottsMPNN: structures are protonation-labelled, featurized into the model’s explicit protonation-token vocabulary, and used to design sequences around pinned protonated centres. The principal workflow runs block-descent redesign, explores stability/selectivity trade-offs, then optionally folds and scores selected designs. Training and structure-labeling are supporting branches; RF3 folding requires external checkpoint weights. The graph follows documented workflows and indexed source references; unsampled implementation details are not inferred.
Main entry points: label_pdb.py PDB labeling · design_ph.py Design workflow · potts_inference.py Potts featurization · fold_rf3.py RF3 folding
Components 19
Protonation labeling
- PDB labeling
labeller/label_pdb.py - His features
labeller/03_his_rich_features.py - Acid features
labeller/04_acid_rich_features.py
PottsMPNN core
- Potts featurization
foundry/models/mpnn/src/mpnn/potts_inference.py - Protonation tokens
foundry/models/mpnn/src/mpnn/transforms/feature_aggregation/token_encodings.py - Graph embeddings
foundry/models/mpnn/src/mpnn/model/layers/graph_embeddings.py - Potts model
foundry/models/mpnn/src/mpnn/model/pottsmpnn.py - Model training
foundry/models/mpnn/src/mpnn/trainers/pottsmpnn.py
pH design
- pH design engine
foundry/models/mpnn/src/mpnn/inference_engines/potts_mpnn_ph.py - Centre placement
inference/design_placement_scan.py - Design workflow
inference/design_ph.py - Pareto selection
inference/design_ph.py
Fold and scoring
- RF3 folding
inference/fold_rf3.py - Charge clashes
scoring/annotate_charge_clash.py - Fold annotation
scoring/annotate_folds.py
Other components
- Designer
- Input structure
- HBPLUS
- RF3 checkpoint
Connections 21
- Designer → PDB labeling: labels structures
README.md - Input structure → PDB labeling: provides PDB
README.md - PDB labeling → HBPLUS: uses geometry
README.md - PDB labeling → His features: labels histidines
- PDB labeling → Acid features: labels acids
- Designer → Design workflow: requests designs
README.md - Input structure → Design workflow: provides backbone
README.md - Design workflow → pH design engine: runs redesign
inference/design_ph.py - Design workflow → Centre placement: scans centres
inference/design_ph.py - pH design engine → Potts featurization: builds context
foundry/models/mpnn/src/mpnn/inference_engines/potts_mpnn_ph.py - Potts featurization → Protonation tokens: encodes tokens
foundry/models/mpnn/src/mpnn/potts_inference.py - Potts featurization → Graph embeddings: featurizes structure
foundry/models/mpnn/src/mpnn/potts_inference.py - Potts featurization → Potts model: runs forward
foundry/models/mpnn/src/mpnn/potts_inference.py - pH design engine → Potts model: optimizes Potts energy
foundry/models/mpnn/src/mpnn/inference_engines/potts_mpnn_ph.py - Design workflow → Pareto selection: selects trade-offs
README.md - Pareto selection → RF3 folding: folds designs
README.md - RF3 folding → RF3 checkpoint: loads checkpoint
README.md - RF3 folding → Charge clashes: scores clashes
inference/design_ph.py - RF3 folding → Fold annotation: annotates folds
inference/design_ph.py - PDB labeling → Model training: supplies labels
README.md - Model training → Potts model: trains model