Theses and Dissertations

Date of Award

5-1-2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

First Advisor

DongChul Kim

Second Advisor

Erik Enriquez

Third Advisor

Emmett Tomai

Abstract

Drug–target affinity (DTA) prediction is a critical step in accelerating drug discovery and reducing its associated costs. Existing computational approaches often employ one-dimensional protein sequences or two-dimensional molecular graphs, which can overlook important three-dimensional (3D) structural features. The present study investigates whether incorporating 3D information of both proteins and ligands can yield more meaningful representations for binding affinity prediction. Specifically, Graph Transformers (GT) are utilized, and the integration of Equivariant Graph Neural Networks (EGNN) is explored to preserve and leverage spatial information during learning.

Using the CrossDocked2020 dataset, which provides protein structures in Protein Data Bank (PDB) format, proteins and ligands are represented as atomic graphs capturing both chemical bonds and geometric relationships. This contrasts with many existing methods relying solely on amino acid sequences for proteins. The proposed framework aims to model the nuanced 3D interactions pivotal to protein–ligand binding by combining attention-based architectures with rotationally and translationally equivariant operations. Potential benefits and challenges of incorporating 3D structural cues in DTA tasks are discussed, along with broader implications for structure-based virtual screening and drug design. Through comparative evaluations of GT and EGNN-enhanced GT, this work provides insights into how 3D information may influence affinity prediction.

Comments

Copyright 2025 Angel Antonio Peredo. https://proquest.com/docview/3240621834

Download

Share

COinS