How huiyum/tac_dose_adjustments works
This repository implements a tacrolimus pharmacokinetic prediction pipeline: patient CSV data is preprocessed, used to train and assess a GRU model, then evaluated on external data and used to produce dose recommendations. The graph follows the visible orchestration and sampled module calls; some cross-validation internals and the contents of app.py were not shown, so uncertain connections are omitted.
Main entry points: main.py Pipeline runner · cross_validation.py Group cross-validation · evaluate.py Prediction evaluation · train.py Model training
Components 17
Pipeline orchestration
- Pipeline runner
main.py - Runtime configuration
src/config.py
Patient data preparation
- Patient CSV data
- Feature preprocessing
src/preprocessing.py - Sequence dataset
src/dataset.py
Model training and assessment
- Group cross-validation
src/cross_validation.py - Model training
src/train.py - GRU PK model
src/model.py - Masked loss
src/losses.py - Prediction evaluation
src/evaluate.py - Evaluation metrics
src/metrics.py
Interpretation and dosing
- Feature importance
src/importance.py - Result visualization
src/visualize.py - Dose optimization
src/dose_optimization.py
Persisted outputs
- Prediction CSVs
- Model artifacts
gru/model.pt
Other components
- Pipeline operator
Connections 26
- Pipeline operator → Pipeline runner: runs
README.md - Pipeline runner → Runtime configuration: reads settings
main.py - Pipeline operator → Patient CSV data: provides
README.md - Pipeline runner → Patient CSV data: reads CSV
main.py - Pipeline runner → Feature preprocessing: preprocesses
main.py - Pipeline runner → Group cross-validation: runs folds
main.py - Group cross-validation → Feature preprocessing: preprocesses folds
src/cross_validation.py - Group cross-validation → Model training: trains folds
src/cross_validation.py - Model training → Sequence dataset: builds batches
src/train.py - Model training → GRU PK model: optimizes
src/train.py - Model training → Masked loss: computes loss
src/train.py - Group cross-validation → Prediction evaluation: evaluates folds
src/cross_validation.py - Prediction evaluation → GRU PK model: predicts with
src/evaluate.py - Prediction evaluation → Evaluation metrics: computes metrics
src/evaluate.py - Group cross-validation → Feature importance: analyzes features
src/cross_validation.py - Feature importance → Sequence dataset: builds permuted data
src/importance.py - Feature importance → Prediction evaluation: measures impact
src/importance.py - Group cross-validation → Result visualization: plots fold results
src/cross_validation.py - Result visualization → Evaluation metrics: uses metrics
src/visualize.py - Pipeline runner → Sequence dataset: builds validation data
main.py - Pipeline runner → Prediction evaluation: evaluates external data
main.py - Pipeline runner → Dose optimization: requests recommendations
main.py - Dose optimization → GRU PK model: predicts with
src/dose_optimization.py - Pipeline runner → Prediction CSVs: writes predictions
main.py - Pipeline runner → Model artifacts: saves weights
main.py - Pipeline operator → Prediction CSVs: receives results
README.md