Gene context drift identifies drug targets to mitigate cancer treatment resistance

Abstract

Cancer treatment often fails because combinations of different therapies evoke complex resistance mechanisms that are hard to predict. We introduce REsistance through COntext DRift (RECODR): a computational pipeline that combines co-expression graph networks of single-cell RNA sequencing profiles with a graph-embedding approach to measure changes in gene co-expression context during cancer treatment. RECODR is based on the idea that gene co-expression context, rather than expression level alone, reveals important information about treatment resistance. Analysis of tumors treated in preclinical and clinical trials using RECODR unmasked resistance mechanisms -invisible to existing computational approaches- enabling the design of highly effective combination treatments for mice with choroid plexus carcinoma, and the prediction of potential new treatments for patients with medulloblastoma and triple-negative breast cancer. Thus, RECODR may unravel the complexity of cancer treatment resistance by detecting context-specific changes in gene interactions that determine the resistant phenotype.

Keywords: DNA repair; cancer; choroid plexus; choroid plexus carcinoma; combination therapy; graph networks; machine learning; radiation; treatment resistance; triple-negative breast cancer.