Machine learning for 3D small molecule drug discovery

Guan, Jiaqi

Permalink

https://hdl.handle.net/2142/124277 Copy

Description

Title

Machine learning for 3D small molecule drug discovery

Author(s)

Guan, Jiaqi

Issue Date

2024-04-15

Director of Research (if dissertation) or Advisor (if thesis)

• Peng, Jian
• Ma, Jianzhu

Doctoral Committee Chair(s)

Peng, Jian

Committee Member(s)

• Banerjee, Arindam
• El-Kebir, Mohammed

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Machine Learning
• Drug Discovery
• Generative Models

Abstract

With the rapid development of geometric machine learning and the availability of ever-increasing biological data, there lies a significant opportunity to expedite drug development processes and substantially reduce associated costs by employing appropriate machine learning (ML) algorithms. This dissertation introduces a suite of tailored ML algorithms aimed at addressing critical challenges in 3D small molecule drug discovery, with the overarching goal of shortening drug development cycles and enhancing drug discovery outcomes. We first investigate the fundamental molecular conformation optimization problem and present a neural energy minimization framework to efficiently and accurately predict molecular conformations. Building upon this groundwork, we extend our framework to atom types and establish connections with diffusion-based generative models. This extension facilitates the introduction of TargetDiff, a SE(3)-equivariant diffusion model to generate ligand molecules for specific protein pockets. We then focus on a specific linker design problem in ROteolysis TArgeting Chimeras (PROTACs) discovery where the fragment poses are unknown, and describe how our proposed LinkerNet addresses this problem with a diffusion model and physics-inspired fragment pose prediction module. Finally, we present a novel paradigm for molecular docking by considering multiple ligands docking to the protein pocket. Collectively, this dissertation showcases the potential of machine learning and deep generative models to revolutionize 3D small molecule drug discovery by translating data into accelerated novel discoveries.

Graduation Semester

2024-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2024 Jiaqi Guan

Owning Collections