UsageNotes_Potency.md

python3 potency_inference.py 
<prompted for options>

Required Inputs

1. Test Dataset (CSV File)

Required columns:

• ligand_smiles (or SMILES, smiles, canonical_smiles) - Chemical structure in SMILES format
• protein_sequence (or PROTEIN_SEQ, protein_seq, sequence) - Amino acid sequence

Optional:

• pIC50 (or pic50, PIC50) - Ground truth binding affinity values (enables metric calculation)

2. Neural Network Model Files

• Model checkpoint (.pt) - Trained GNN or GPFT model weights
• Vocabulary (.pkl) - Amino acid to index mapping
• Tokenizer (.pkl) - Protein sequence tokenizer

3. XGBoost Model Files

• XGBoost model (.json or .pkl) - Trained gradient boosting model
• Feature scaler (.pkl) - StandardScaler for descriptor normalization
• Descriptor list (.txt) - Names of RDKit molecular descriptors
• Docking scores CSV (optional) - Pre-computed docking scores
  • Columns: ligand_smiles, protein_sequence, docking_score

4. Stacking Model File

• Ridge regression model (.pth) - Meta-learner that combines predictions

5. User Selections (Interactive)

• Model type: GNN or GPFT
• Split strategy: Random or Scaffold (must match training)
• If XGBoost model uses docking scores

Generated Outputs

Output Directory Structure:

predictions/ (or custom name)
├── test_predictions.csv
├── metrics.json
├── config.json
└── predictions_plot.png