Stay Informed:

COVID-19 (coronavirus) information
Zoom Links: Zoom Help | Teaching with Zoom | Zoom Quick Guide

Defense: Prediction of cancer phenotypes through the integration of multi-omic data and prior information

Speaker Name: 
Vladislav Uzunangelov
Speaker Title: 
PhD Candidate (Advisor: Josh Stuart)
Speaker Organization: 
Biomolecular Engineering & Bioinformatics
Start Time: 
Friday, May 10, 2019 - 1:00pm
End Time: 
Friday, May 10, 2019 - 3:00pm
Location: 
Physical Sciences Building, Room 305
Organizer: 
Josh Stuart

Abstract:  High-throughput data have become ubiquitous in the study of biological phenomena. We can now query cellular state at higher resolution, giving us better insight into complex diseases. For example, there are currently tens of thousands of cancer patients with simultaneous copy number, mutation, methylation, mRNA, miRNA and protein level profiles. Furthermore, experimental cellular perturbations are increasingly compared on the multi-omic level. Such experiments uncover important dependencies among genes, their products and environmental conditions - relationships that accumulate in a growing number of databases. However, the integration of this prior pathway knowledge with new heterogeneous genomic measurements in an interpretable model remains a formidable challenge that is still not fully solved.

I present two kernel-based approaches to tackle that problem. The first one uses pathway-based kernel functions in a Multiple Kernel Learning framework to jointly win the DREAM9 Gene Essentiality Prediction Challenge. The second improves upon the DREAM9 winner by introducing empirical kernel functions derived from Random Forest tree ensembles. The latter is shown to outperform state-of-the-art methods in diverse phenotype learning tasks, including predicting microsatellite instability in endometrial and colorectal cancer, survival in breast cancer and shRNA knockdown response in CCLE cell lines. Finally, I demonstrate how the Algorithm for Kernel Learning with Approximating Tree Ensembles (AKLIMATE) can be adapted to the development of multi-omic minimum-feature predictors for patient subtypes.