Date of Award

Fall 1-1-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Chemistry

First Advisor

Batista, Victor

Abstract

At the intersection of quantum computing and machine learning, quantum machine learning (QML) has the potential to drive major breakthroughs in chemistry and drug discovery. This dissertation explores the development of quantum neural networks and their application within drug discovery tools, navigating a deliberate path from current methodologies towards quantum enhancement under progressively more practical assumptions. Initially, we present novel Quantum Convolutional Neural Network (QCNN) architectures for multi-channel supervised learning (Chapter 2), laying the groundwork for use in a wide variety of drug discovery tasks such as protein-ligand binding affinity and structure prediction. Despite state-of-the-art performance among QCNNs on benchmark, no quantum speed-up is reported. To remedy this, Chapter 3 introduces a QCNN framework for drug toxicity prediction that leverages the Hadamard test to achieve a theoretical quadratic speedup in the critical inner product calculation between input data and kernel. This advancement, however, relies on the assumption of efficiently prepared quantum states. Addressing this dependency, Chapter 4 transitions to generative modeling, presenting a hybrid quantum-classical Transformer architecture for property-conditioned molecular generation. This model adapts Chapter 3's concepts of efficient quantum inner products for its self-attention mechanism but implements them without requiring the assumption of efficiently prepared states. This work discusses the theoretical foundations of QML, including data encoding, variational quantum circuits, and hybrid quantum-classical approaches, and culminates in a discussion of future pathways for integrating these quantum advancements into broader chemical and pharmaceutical research (Chapter 5 & 6).

Download

Share

COinS