Computational modeling of thousands of tumors

Through large international collaborations, thousands of clinical tumor samples have already been profiled using standard  “bulk” methods. We use gene expression signatures of subpopulations derived from our single cell studies to re-analyze those bulk profiles and model them as a mixture of diverse subpopulations (Science 2016). This modeling enables us to predict the composition of each tumor, define subtypes of tumors, and infer cell-to-cell interactions. We identify tumors enriched with particular subpopulations and associate those subpopulations with clinical features such as response and resistance to treatments, metastasis and survival.