CANCERCELLNET: A COMPUTATIONAL PLATFORM TO EVALUATE THE TRANSCRIPTIONAL FIDELITY OF CANCER MODELS

Cancer cell lines, patient derived xenografts and genetically engineered mouse models provide valuable platforms for researchers to investigate tumor biology and to identify therapies. The extent of similarity between cancer models and native tumor is crucial for translational research but often unclear. With the preponderance of cancer models and the ability to generate new ones more readily, there is a demand for a computational tool that can readily assess the extent of similarity between newly generated cancer models and native tumors. In this thesis, we address the two concerns by first developing a computational tool, CancerCellNet, that can robustly measure transcriptional similarities between cancer models and tumor samples from 22 tumors types and 36 cancer subtypes in a platform and species agnostic manner. Secondly, to address the extent of similarities between cancer models and native tumors, we applied CancerCellNet to 657 cancer cell lines, 415 patient derived xenografts and 26 distinct genetically engineered mouse models, documenting the most faithful models, identifying cancer categories lacking adequate models and finding models with annotations that do not match their classification. By comparing models across modalities, we found that genetically engineered mice have higher average transcriptional fidelity than patient derived xenografts and cell lines in four out of five tumor types. We provided our analysis of cell lines, patient derived xenografts, and genetically engineered mouse models as a web application for researchers to select the most representative models for their research question. We have also made CancerCellNet available as an opensource software and as a web application that can be used to assess new cancer models