Abstract
Cancer cells depend on multiple driver alterations whose oncogenic effects can be suppressed by drug combinations. Discovery of effective combination therapies is challenging due to the complexity of the biomolecular landscape of drug responses. Here, we developed the method REFLECT (REcurrent Features Leveraged for Combination Therapies), which integrates machine learning and cancer informatics algorithms. The method maps recurrent co-alteration signatures from multi-omic data across patient cohorts to combination therapies. Using the REFLECT framework, we generated a precision therapy resource matching 2,201 drug combinations to co-alteration signatures across 201 cohorts stratified from 10,392 patients and 33 cancer types. We validated that REFLECT-predicted combinations introduce significantly higher therapeutic benefit through analysis of independent data from comprehensive drug screens. In patient cohorts with immunotherapy response markers, HER2 activation and DNA repair aberrations, we identified therapeutically actionable co-alteration signatures shared across patient sub-cohorts. REFLECT provides a framework to design combination therapies tailored to patient cohorts in data-driven clinical trials.
Competing Interest Statement
The authors have declared no competing interest.