AI for precision oncology

Abstract

This talk centers around the design and implementation of robust, end-to-end machine-learning pipelines for predicting patient response from high-dimensional molecular data. We’ll walk through each critical stage—data ingestion and normalization, leakage-proof nested cross-validation, stability-based feature selection to derive reproducible biomarker panels, hyperparameter optimization, calibration of probability outputs, and external cohort validation—highlighting why every step matters for building reliable, generalisable models.