Mapping the molecular landscape to optimal treatments

What is the best treatment for each patient?

In this theme we developed a number of approaches to map molecular profiles of tumors to optimal treatments. These approaches include 1) supervised approaches for finding predictive biomarkers, i.e. biomarkers that predict response to therapy; 2) approaches that use an unsupervised clustering of tumor samples as a starting ‘landscape’ and then map these clusters to treatments via molecularly and chemically profiled cell line panels as well as 3) approaches that rely on in vitro studies to identify key modulators of disease progression and resistance and use gene expression signatures to identify subgroups of patients that exhibit these same characteristics and would therefore potentially benefit from the treatments targeting the key regulators. When companion treatments for the regulators are not known, the signatures are employed to identify corresponding cell line populations, and drugs showing the greatest differential response in these cell lines are candidate treatment options.


  • Molecular subtyping of cancer types
  • Identifying subgroup markers in heterogeneous populations
  • Detecting markers of resistance
  • Representativity of different models (cell lines, tumor samples, organoids)
  • Predicting drug response based on the molecular landscape
  • Identifying drug synergy and associated molecular biomarkers


  • iNMF
  • DIDS
  • logic models