An ensemble based approach for drug sensitivity prediction

Drug sensitivity prediction based on genomic characterization remains a significant challenge in the area of systems medicine. Multiple approaches have been proposed for mapping genomic characterization to drug sensitivity and among them ensemble based learning techniques such as random forests have turned out to be a top performer. In the first part of this thesis, we consider the problem of predicting sensitivity of cancer cell lines to new drugs based on supervised learning on genomic profiles. The genetic and epigenetic characterization of a cell line provides observations on various aspects of regulation including DNA copy number variations, gene expression, DNA methylation and protein abundance. To extract relevant information from the various data types, we applied a Random Forests based approach to generate sensitivity predictions from each type of data and combined the predictions in a linear regression model to generate the final drug sensitivity prediction. Our approach when applied to the NCI-DREAM drug sensitivity prediction challenge was a top performer among 47 teams and produced high accuracy predictions. Our results show that the incorporation of multiple genomic characterizations lowered the mean and variance of the estimated bootstrap prediction error. We also applied our approach to the Cancer Cell Line Encyclopedia database and it produced high accuracy drug sensitivity prediction with the ability to extract the top targets of an anti-cancer drug. The results illustrate the effectiveness of our approach in predicting drug sensitivity from heterogeneous genomic datasets. For the purpose of further exploring the predictability of anti-cancer drug sensitivities, we observe that majority of current approaches infer a predictive model for each drug individually, but correlation between different drug sensitivities suggests that multiple response prediction incorporating the co-variance of the different drug responses can possibly improve the prediction accuracy. In the second part, we present a prediction and analysis framework based on Multivariate Random Forests that incorporates the correlation between different drug sensitivities. The results of application of our framework to the Genomics of Drug Sensitivity in Cancer project dataset and Cancer Cell Line Encyclopedia data shows marked improvement over regular Random Forests and Elastic Net. The presented framework was also utilized to generate multivariate probability distributions of the predicted output responses. Experimental results show that conditional expectation based on multivariate probability distribution and knowledge of the response of a correlated drug can be utilized to considerably improve prediction accuracy.

Name thesis:degree_name

Master of Science in Electrical Engineering

Level thesis:degree_level

Masters

Discipline thesis:degree_discipline

Electrical Engineering

Grantor

Texas Tech University

Year dc:date.issued

2014

Author dc:creator

• Wan, Qian

Chair dc:contributor.committeechair

• Pal, Ranadip

Committee members dc:contributor.committeemember

• Nutter, Brian
• Roeger, Lih-Ing W.

• Random Forest
• Multivariate Random Forest
• Drug Sensitivity Prediction

Language dc:language.iso

eng

Handle dc:identifier.uri

https://hdl.handle.net/2346/89033

OAI identifier oai:identifier

oai:ttu-ir.tdl.org:2346/89033