Genomics-Guided Discovery of Effective Combination Therapies in Cancer

Development of combination therapies guided by genomic data is a highly promising precision medicine strategy. As a team of bioinformatics experts and clinician scientists, we are combining cancer genomics and machine learning algorithms to accelerate the discovery of effective combination therapies in diverse cancer types. With the help of the ICI fund, we are building an integrated bioinformatics pipeline for computational identification of co-occurring genomic aberrations that can be targeted with drug combinations.

Our goal is to accelerate discovery of combination therapies through computational analysis of genomic and proteomic data.

Multiple oncogenic processes are co-activated in individual tumors due to co-occurring oncogenic aberrations. Precision medicine strategies with mono-agent interventions usually fail to suppress the co-activated pathways.   There is an unmet need for genotype specific therapeutic strategies that target the co-activated oncogenic processes.

The nature of the co-activated oncogenic pathways varies from patient to patient, and yet shows recurrent characteristics since they introduce selective advantages to tumor cells in terms of response to therapy and increased growth rates. Therefore, we hypothesize that combination therapies that target recurrently co-activated pathways can induce more effective and durable responses compared to monotherapies. 

Research Strategy
To test this hypothesis and accelerate discovery of biomarker-driven combination therapy, we have established a computational/experimental prediction and validation platform (figure 1). We follow a two-step strategy that involves; (i) selection of combination therapy targets through bioinformatic analysis of large-scale genomic data using a feature selection/clustering based statistical algorithm, (ii) a bioinformatics-guided drug combination screen to validate the predicted combination therapy strategies. The outcome of the computational analysis is a small set of recurrent, actionable features that co-occur in tumor sub-cohorts and discriminate the samples in a sub-cohort from the rest.

Progress Report 1

We completed the pilot phase of the project supported by ICI. We conclude that the project has provided a framework for accelerated discovery of combination therapies from genomic and gene/protein expression data. As the combination precision therapy trials such as NCI combination MATCH is emerging, our innovation and resulting product, REFLECT will guide clinicians to make better therapy choices driven by data.

  • We integrated genomic (mutation and copy number) and proteomics data from patients (TCGA project and other sources) and cell lines (CCLE, Sanger GDSC and MD Anderson cell line project).
  • We implemented the feature selection/clustering algorithm for identification of co-occurring aberrations.
  • Using TCGA mutation, copy number and proteomics data, we identified co-occurring features in distinct classes of cancer types marked with actionable genomic aberrations
  • We identified co-actionable and co-occurring aberrations across patient cohorts
  • We validated the identified drug combinations in cell lines mathicing to the genomic or proteomic signatures from REFLECT calculations
  • We are currently supporting clinical trials and research at MD Anderson Cancer Center as we analyze the distribution and co-occurrence patterns of genomic alterations that confer resistance to immunotherapy resistance.
  • A manuscript is in progress to disseminate our results from this project.
  • All of our analysis is currently shared in a pilot R-Shiny platform:
  • We are moving to a more stable and scalable portal supported by MDACC IT team:

Final Report

  1. We have implemented the REFLECT precision combination therapy portal at:
    The data portal is 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.
  2. The REFLECT source code and package for the machine learning algorithm is shared at: This is an open-source bioinformatics/machine learning tool.
  3. The manuscript is currently in review. The preprint is available at: