How bio-ontology-research-group/d4 works
D4 is a research workflow for predicting drug–drug interactions and interpreting them across 11 biological mechanisms. Its principal path uses drug embeddings and labeled interaction data to tune a neural network, trains/evaluates a model, ranks candidate interactions, and records evaluation data for AUC plotting. The README documents the handoff from hyperparameter search to ANN training. The scripts use hard-coded local file paths; their external data locations and any additional mechanism-processing stages are not implemented in this tree.
Main entry points: ANNmodel.py ANN training · Hypers.py Hyperparameter search · AUC-code.py AUC analysis
Components 8
Model inputs
- Drug embeddings
- DDI labels
Modeling and prediction
- Hyperparameter search
Hypers.py - ANN training
ANNmodel.py - Interaction ranking
ANNmodel.py
Evaluation
- Rank results
ANNmodel.py - AUC analysis
AUC-code.py - AUC plot
AUC-code.py
Connections 9
- Drug embeddings → Hyperparameter search: provides vectors
README.md - DDI labels → Hyperparameter search: provides labels
Hypers.py - Hyperparameter search → ANN training: guides parameters
README.md - Drug embeddings → ANN training: provides vectors
ANNmodel.py - DDI labels → ANN training: provides labels
ANNmodel.py - ANN training → Interaction ranking: scores candidates
ANNmodel.py - Interaction ranking → Rank results: writes ranks
ANNmodel.py - Rank results → AUC analysis: loads rank data
AUC-code.py - AUC analysis → AUC plot: plots curves
AUC-code.py