Interpretable classification of molecular measurements via pathway-induced multiple kernel learning

Roadmap

Molecular data classification: use network topology as a regulariser, define meta-features using pathway information Pathway-Induced. Multiple Kernel Learning (PIMKL): concept, pathway induction, multiple kernel learning PIMKL benchmarking: benchmark against other prior knowledge informed methods on multiple breast cancer cohorts.
PIMKL application: detects tumor samples accurately, integrates multiple omics seamlessly, identifies relevant pathways for each data type.