Bilkent University
Department of Computer Engineering


DeepSide: A Deep Learning Framework for Drug Side Effect Prediction Using Gene Expression Signatures


Onur Can Üner
MS Student
Computer Engineering Department
Bilkent University

Drug Adverse drug reactions (ADRs) are unintended responses to a drug. Unreported ADRs can detriment human health and are the main reason for the withdrawal of a drug from the market; therefore, prediction of ADRs at the clinical trial phase is desirable. Traditional methods aim to predict the ADRs of a given drug by solely relying on its chemical and biological properties. A recent work, on the other hand, utilized the gene expression profiles of the cells in addition to chemical properties of the drug. For this purpose they made use of the gene expression dataset obtained in the Integrated Network-based Cellular Signatures (LINCS) L1000 project which profiled gene expression changes in different human cell lines upon treating them with a large panel of drugs and small-molecule compounds. The best performing method in that study was an extreme-tree classifier. In this study, we aim to improve upon the current state-of-the-art by through deep learning that leverage the LINCS data more effectively and unlike current approaches, we account for semantic relationships that relate labels such as “abdominal pain” and “abdominal cramp”. Our results show that the presented method offers measurable improvements over the current-state-of-the-art tools.


DATE: 02 April, 2018, Monday, CS590 & CS690 presentations begin at @ 15:40