Show simple item record

dc.contributor.advisorLeslie, Christina
dc.contributor.authorYang, Li
dc.description.abstractBoth innate and acquired resistance to molecularly targeted therapies represent major challenges to cancer treatment. Recent studies have proposed a role for the tumor microenvironment in therapeutic response, demonstrating that cytokines secreted by stromal cells can rescue cancer cells from killing by targeted drugs. To systematically study the stromal contribution to innate drug resistance, we used a method called affinity regression to model the effect of stromal cells on cancer cell drug sensitivity using a large published stromal-cancer co-culture data set. Our model represents each stromal cell by the feature vector of expression levels of its secreted cytokines, and each cancer cell by pathway scores derived from curated signaling pathway databases, giving a view of the cellular circuitry that could receive and transduce signals from stromal cells. For each drug, our algorithm trains a regularized bilinear regression model that predicts the stromal rescue score for a cancer cell line from stromal and cancer cell features. We confirmed that affinity regression outperformed nearest neighbor approaches for the task of predicting rescue scores in cross-validation experiments. Furthermore, by analysis of the trained model, we identified cytokines secreted by stromal cells that may interact with signaling pathways in cancer cells to mediate rescue. For the BRAF inhibitor PLX4720, our model identified HGF as the cytokine most predictive of melanoma cancer cell rescue and associated with c-MET and PI3K signaling, consistent with published experimental reports. Our model also predicted that HGF plays a similar role in non-small cell lung carcinoma (NSCLC) cells treated with EGFR inhibitors, and we confirmed this prediction experimentally for afatinib and erlotinib. Our statistical model of tumor-stromal interactions may lead to new insights into the role of stromal cells in promoting drug resistance and could ultimately suggest combination therapies to target the tumor microenvironment.
dc.rightsAttribution-NonCommercial-NoDerivatives 4.0 International
dc.subjectdrug resistance
dc.subjectmachine learning
dc.subjecttumor microenvironment
dc.titleModeling Biological Interactions Using Supervised Machine Learning
dc.typedissertation or thesis Biology and Medicine Cornell Graduate School of Medical Sciences of Philosophy

Files in this item


This item appears in the following Collection(s)

Show simple item record

Except where otherwise noted, this item's license is described as Attribution-NonCommercial-NoDerivatives 4.0 International