Cancer cells respond to the signaling molecules they encounter in different ways, and these differences play an important role in predicting tumor development and drug response. Breast cancer cell lines are traditionally divided into three subtypes, but the existing classification may not be a good predictor of response to signaling molecules.
We develop a new method to group the responses which keeps track of the multi-dimensional structure of the original data. With imposed “interpretability” restrictions on the final clusters, we proceed via integer programming. The final stage is to construct mechanistic models that summarize, for each cluster, the crosstalk between two main communication pathways AKT and ERK.