Doctor of Philosophy (PhD)


Biological Sciences

Document Type



Recent advances in the analysis of omics data from various cancer cells pave the way for cancer therapy to predict effective drug responses and drug-target interactions based on cell- specific genetic features such as gene expression profiles, mutations, and copy number variations. Because the phenotype of a complex disease, such as cancer, is determined by several factors, it is difficult to predict the best effective treatment for a certain cell line based just on one sort of data, such as genetic traits and drug molecular features. The overall studies in this thesis proposed a graph-based representation of heterogeneous biological data and applications of this data representation in graph-based deep learning approaches that can use network data alongside other types of data to predict effective drug response to account for the complexities of cancer progression.

The overall thesis has been divided into four sections. In the first section, we addressed the question that how to integrate multiple heterogeneous data to represent the system-level complexity as a graph in the study of anti-cancer pharmacotherapy. In the second section, we suggested the implementation of a graph-based artificial intelligence system to predict drug reactions, mostly kinase inhibitors, on diverse cancer cell lines. In the third section, we compared our proposed method with gene signature analysis and validated our results against biomedical literature and live-cell time course inhibition assays. In the concluding part, we suggested an artificial intelligence technique for predicting drug-target interactions based on graph-based data.



Committee Chair

Brylinski, Michal



Available for download on Friday, July 11, 2025