Transfer learning with multiomics integration and deep neural networks reveals drug resistance mechanisms in cancer

Description

Semih Alpsoy and Osman Ugur Sezerman, publishing in 2025, use deep learning to predict when cancer drugs will fail, and why. Drug resistance is one of the main reasons cancer treatment falls short. The team trained neural networks on cell-line data that combines gene activity, mutations, and copy number changes with measured drug responses. They then borrowed strength across related drugs through transfer learning.

The models beat rival methods at predicting responses in two independent test sets, one from patient-derived tumors and one from a large cancer atlas. Digging into four drugs, the team flagged specific pathways behind resistance. A metabolism route and a cell-grip route stood out for one drug, and a cleanup process for another. Some resistance links also sat in the copy number data. The authors note their findings match earlier lab studies, which supports the approach.

Catalog record

Source: https://europepmc.org/articles/PMC12661044