Bilkent University
Department of Computer Engineering

---

M.S.THESIS PRESENTATION

 

A Graph-based Learning Model for Predicting Drug Combination Synergies

 

Muammer Buğra Kurnaz
Master Student
(Supervisor: Assoc.Prof. A.Ercüment Çiçek)

Computer Engineering Department
Bilkent University

Abstract: Drugs can treat patients more effectively when used in combination rather than as a single agent. Biologists conduct manual experiments to learn which drugs perform better when combined, yet this process is tedious and costly. Studies have revealed that machine learning methods may help with the discovery of synergistic drug combinations. In this thesis, we present a GNN-based architecture for drug synergy prediction. This model achieves improved predictions in settings where either one or both drugs are absent from the training set, which is a limitation of most drug synergy prediction models. To evaluate generalization to unseen drugs more rigorously, we introduce a degree-stratified splitting algorithm based on drug co-occurrence graphs. The proposed model improves the Pearson correlation in this setting from 0.107 to 0.228, and Spearman correlation from 0.151 to 0.245, averaged across three independent splits.

 

DATE: June 02, Tuesday @ 11:45 Place: EA 409