Theses and Dissertations

Date of Award

8-1-2026

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

First Advisor

Dongchul Kim

Second Advisor

Erik Enriquez

Third Advisor

Haoteng Tang

Abstract

The discovery of novel drug candidates remains a significant challenge due to the vast size of chemical space and the need to simultaneously optimize multiple objectives, including molecular quality, novelty, diversity, and computational efficiency. This thesis investigates the design and evaluation of heuristic-guided search strategies for de novo molecular generation. Four search algorithms, Best-First Search, A* Search, Monte Carlo Tree Search (MCTS), and Genetic Algorithms (GA), were evaluated in combination with four heuristic functions incorporating molecular property optimization, novelty promotion, and search-progress information. Candidate molecules were generated and assessed using RDKit-based metrics, including drug-likeness, diversity, uniqueness, and novelty relative to a reference dataset. A comprehensive experimental study was conducted across multiple search budgets and random seeds to evaluate the effectiveness, efficiency, and robustness of each algorithm–heuristic combination. The results demonstrate that heuristic design plays a critical role in shaping search behavior and determining the quality of generated molecules. Property-driven heuristics consistently produced candidates with strong objective scores, whereas novelty-aware heuristics encouraged broader exploration of chemical space and generated more diverse molecular structures. Furthermore, the findings reveal meaningful interactions between search algorithms and heuristic formulations, highlighting inherent trade-offs between optimization performance and molecular diversity. This work presents a systematic framework for evaluating heuristic-guided molecular generation approaches and provides empirical insights into how different search strategies navigate chemical space. The results contribute to the development of more efficient, interpretable, and effective computational methods for early-stage drug discovery. In addition, the proposed framework establishes a foundation for future research integrating machine learning and reinforcement learning techniques into molecular design workflows, with the goal of improving both search efficiency and the quality of generated drug candidates.

Comments

Copyright 2026 Evans Kwame Sarkodie. All Rights Reserved. https://proquest.com/docview/3408996812

Download

Share

COinS