How sarhnagniafariba-maker/tnbc-immunotherapy-prediction works
This is a research prototype for predicting TNBC immunotherapy response from multi-omics and clinical inputs. Its intended workflow is cohort data preparation, graph construction, GCN training, held-out evaluation, and SHAP/attention-based interpretation. The graph preserves those documented stages and the sampled training/model/analysis modules. Several documented capabilities (including GEO integration and preprocessing) are not represented by dedicated visible files, so edges to them are limited to what the README or excerpts establish; the README also describes a future rather than implemented multimodal scope.
Main entry points: train.py Fold preprocessing and training · evaluate.py Predictive evaluation · explain.py SHAP and ablation analysis · gcn-models.py GCN predictor
Components 11
Cohort data
- Synthetic cohort generator
data/make-dummy-data.py - GEO and cohort inputs
- Feature graph builder
utils/graph-utils.py
Graph learning
- Fold preprocessing and training
script/train.py - GCN predictor
models/gcn-models.py
Model analysis
- Predictive evaluation
script/evaluate.py - Metric utilities
utils/metrics.py - SHAP and ablation analysis
script/explain.py - Response predictions
- Biomarker interpretations
Other components
- Researcher
Connections 16
- Researcher → Synthetic cohort generator: generates test data
data/make-dummy-data.py - Synthetic cohort generator → GEO and cohort inputs: writes synthetic inputs
data/make-dummy-data.py - GEO and cohort inputs → Fold preprocessing and training: loads inputs
script/train.py - Researcher → Feature graph builder: builds feature graph
utils/graph-utils.py - Feature graph builder → Fold preprocessing and training: provides adjacency
- Fold preprocessing and training → GCN predictor: instantiates and trains
script/train.py - GCN predictor → Fold preprocessing and training: returns predictions
models/gcn-models.py - Fold preprocessing and training → Response predictions: produces scores
script/train.py - GEO and cohort inputs → Predictive evaluation: loads inputs
script/evaluate.py - Predictive evaluation → GCN predictor: loads trained model
script/evaluate.py - Predictive evaluation → Metric utilities: calls DeLong test
script/evaluate.py - Predictive evaluation → Response predictions: reports metrics
script/evaluate.py - GEO and cohort inputs → SHAP and ablation analysis: loads inputs
script/explain.py - SHAP and ablation analysis → GCN predictor: loads and queries
script/explain.py - SHAP and ablation analysis → Metric utilities: imports metric symbol
script/explain.py - SHAP and ablation analysis → Biomarker interpretations: saves explanations
script/explain.py