INTEGRATION AND ANALYSIS OF BIOLOGICAL DATA FOR COMPUTATIONAL DRUG DISCOVERY

Date

2023-6-05

Author

Ataş Güvenilir, Heval

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

430
views

1374
downloads

Cite This

Drug discovery and development is a slow and costly process that comprises identifying bioactive compounds against biomolecular targets and evaluating their efficacy and safety. Computational drug/compound–target/protein interaction (DTI/CPI) prediction approaches have emerged as valuable tools to streamline this process and minimize expenses. In recent years, the integration of artificial intelligence (AI) based methods in DTI prediction has gained considerable attention, but challenges persist due to limitations in existing approaches and the complex nature of this biological problem. This thesis study aims to contribute to the effective utilization of AI in drug discovery by addressing current obstacles and developing innovative DTI prediction models. The main goal is to establish a reliable standard for designing robust and industry-applicable computational systems. The study is divided into three parts, each addressing a different aspect of the problem. In the first part, we performed a comprehensive benchmark for machine learning-based DTI prediction to achieve better data representations and more successful learning, and proposed high-quality bioactivity datasets for a fair and reliable comparison. In the second part, we utilized the knowledge graph (KG) data structure to leverage heterogeneous biological data for improved drug discovery, and constructed the KG module of our biological data integration system (CROssBAR) by incorporating essential relationships among multiple types of biomedical entities. In the last part, we proposed HetCPI, a systems-level CPI representation and prediction framework, which utilizes cutting-edge heterogeneous graph representation learning algorithms to extract hidden knowledge from multi-layered biomedical data, i.e., CROssBAR KGs, and demonstrates a considerable performance improvement in challenging scenarios. The outputs of this thesis study are expected to aid experimental and computational work in biomedical sciences, especially in drug discovery and repurposing.

Subject Keywords

machine/deep learning, bioactivity modeling, protein representation, biomedical knowledge graph, heterogeneous graph representation learning

URI

https://hdl.handle.net/11511/104453

Collections

Graduate School of Informatics, Thesis