How patrickbryant1/umol works
Main Components
a. Input Processing Layer:
• Multiple Sequence Alignment (MSA) processing
• SMILES string processing for ligands
• Protein pocket information processing
b. Core Neural Network Architecture:
• Feature extraction modules
• Common modules layer
• Folding modules
• Model components from src/net/model/
• This as the central processing unit
c. Output Processing Layer:
• Structure prediction output
• Relaxation pipeline
• Visualization components
Component Relationships
• Data flow from input processors to the neural network
• Internal connections between network modules
• Flow from network to structure prediction
• Connection to the relaxation pipeline
Specific Implementation Details
Input Processing:
• MSA Features (src/make_msa_seq_feats.py)
• Ligand Features (src/make_ligand_feats.py)
• Pocket Processing (src/parse_pocket.py)
Core Network (src/net/):
• Model Configuration (model/config.py)
• Feature Processing (model/features.py)
• Core Model (model/model.py)
• Folding Logic (model/folding.py)
Output Processing:
• Structure Prediction (src/predict.py)
• OpenMM Relaxation (src/relax/openmm_relax.py)
Technical Details
• CUDA dependency
• OpenMM integration
• Data format transformations
• Model parameter handling