greenelab/pancancer

Implements elastic net logistic regression classifiers trained on multi-modal TCGA data (gene expression, mutations, and copy number alterations) to detect pathway activation and tumor suppressor inactivation across cancer types. The modular pipeline uses MAD-based feature selection and cross-validation with hyperparameter grid search, allowing flexible classifier construction for any gene set and disease combination through command-line configuration. Includes pre-built analyses for TP53 and Ras pathway dysregulation with published validation across phenocopying events like NF1 loss of function.

122 stars. No commits in the last 6 months.