How kochgroup/famdlike works
FamDLike is a family-specific molecular drug-likeness scorer. Its main workflow accepts molecular records, optionally derives RDKit descriptors from SMILES, then produces family desirability scores and optional GAM and tree-ensemble probabilities for a composite score and class prediction. Separate experiment pipelines train the three scoring components. The README documents their roles and outputs; sampled source confirms descriptor calculation and model interfaces, but does not expose every inference call or artifact-loading detail, so those links are cited to documentation where appropriate.
Main entry points: main.py Prediction API / CLI · optimizer.py Property distribution fitting · desirability_functions.py Desirability functions · des_engine.py FDS engine training
Components 14
Prediction interface
- Prediction API / CLI
main.py - Feature configuration
config.py
Molecular features
- RDKit descriptors
calc_desc.py
Family desirability
- FDS engine training
desirability_functions/des_engine.py - Property distribution fitting
desirability_functions/optimizer.py - Desirability functions
desirability_functions/desirability_functions.py - Fitting utilities
desirability_functions/des_fitting.py
GAM scoring
- GAM training
gam/gam_training.py - GAM model
gam/gam_model.py
Tree ensemble
- Tree model training
tree_models/training_treemodels.py - Tree classifiers
tree_models/treemodels.py - Training feature filters
tree_models/utils.py
Other components
- Molecule user
- Scored results
Connections 14
- Molecule user → Prediction API / CLI: submits molecules
README.md - Prediction API / CLI → Feature configuration: loads features
main.py - Prediction API / CLI → RDKit descriptors: calculates descriptors
main.py - Prediction API / CLI → Desirability functions: scores FDS
README.md - Prediction API / CLI → GAM model: predicts probabilities
README.md - Prediction API / CLI → Tree classifiers: uses ensemble
README.md - Prediction API / CLI → Scored results: returns scores
README.md - FDS engine training → Property distribution fitting: fits distributions
desirability_functions/des_engine.py - FDS engine training → Desirability functions: builds engine
desirability_functions/des_engine.py - Property distribution fitting → Desirability functions: configures functions
desirability_functions/optimizer.py - Fitting utilities → Property distribution fitting: uses fitting
desirability_functions/des_fitting.py - GAM training → GAM model: fits GAM
README.md - Tree model training → Tree classifiers: trains classifiers
tree_models/training_treemodels.py - Tree model training → Training feature filters: uses filters
tree_models/training_treemodels.py