DRUG TARGET DECONVOLUTION IN CANCER CELL LINES

Abstract

The deconvolution problem to identify the critical protein targets behind drug sensitivity profiling is an important part of drug development. It helps us to understand the mechanism of action of anti-cancer drugs on the cell lines through protein targets in those cell lines. This problem can be formulated as a matrix deconvolution problem, with two matrices for the cell-based drug sensitivity profiling and drug target interaction data, respectively. The model needs to be solved to identify the vulnerability of the cell lines to inhibition of critical targets. We used drug sensitivity data for 265 anti-cancer compounds over 990 cell models taken from cancer patients and cultivated in the lab. Using the data on interaction of these drugs with the protein targets, we used a novel method called TDSBS (target deconvolution with semi-blind source separation) in order to determine the critical targets for each cell model. The critical protein targets determined using this method were found to be clinically relevant, as we could determine that the driver genes have higher TDSBS values compared to the non-driver genes in the cell models. In this thesis we demonstrate a general statistical model which can be used to identify the protein targets which are inhibited by anti-cancer drugs in drug/cell line sensitivity experiments.