Sikander Hayat, Ph.D.
Visiting Research Scientist
Broad Institute of MIT and Harvard
Selecting the next top (oncology) model – refining cancer sub-types and selecting best in vitro models based on machine learning using multi-omics data
Cancer is a heterogeneous disease with well-established sub-types that have different genomic properties and are associated with different outcomes. As such, information about sub-type specific information can be used to identify potential sub-type specific therapies. Additionally, genomic correlations in these sub-types can be exploited clinically by developing robust patient stratification biomarkers. However, in order to identify novel dependencies and targeted therapies, it is essential to identify pre-clinical models such as cell lines, xenografts and patient derived cells that best represent the varying genomic landscape of patient populations. In this study, we focus on colorectal cancer and use a multi-omics deep-learning approach to refine existing colorectal cancer sub-types seen in patient samples in The Cancer Genome Data (TCGA) data and identify cell lines that best represent colorectal cancer sub-types in an unbiased manner. In addition, we identified robust biomarkers for each sub-type and developed a classifier to automatically assign best representative cell lines to any patient sample with adequate genomic data available. These findings are relevant for patient stratification and selection of cell lines for early stage drug discovery pipelines, biomarker discovery and target identification.
Support for this seminar is provided by an Institutional Development Award (IDeA) from the National Institute of General Medical Sciences of the National Institutes of Health under grant number P20GM103423.