Abstract
dc:descriptionDrugs, through their interaction mainly with specific target proteins, regulate biochemical pathways and physiological functions in order to prevent and treat various diseases. The increasing costs and long development periods of new drugs are a major challenge for the pharmaceutical industry. According to recent data, 92.5% of drug candidates fail. This leads to a significant loss of resources and time with the average total cost of developing a drug being $879.3 million and the average development time being 15 years. The cost of developing new drugs is expected to continue to rise in the future, making it imperative to find new approaches that will reduce the cost and time of development. In recent years, machine learning has made significant contributions in this direction. Moreover, the large increase in available biological data has helped deep learning methods to demonstrate superior performance compared to conventional machine learning and other traditional computational methods. However, these methods are complex and time-consuming, which is a major obstacle to their practical application. In this context, quantum computing and quantum machine learning offer theoretically promising solutions to effectively address these challenges. This thesis addresses classical machine learning and quantum computing, and how these fields can be effectively combined to create a new research field of quantum machine learning. Furthermore, this paper presents in detail the applications of quantum machine learning in the field of drug discovery. Predicting the affinity of drug binding to target proteins is one of the most important issues in drug discovery. We present classical machine learning algorithms that predict binding affinity and discuss various issues such as data representation. We propose hybrid quantum-classical neural networks for predicting the binding affinity of target protein and potential drug as well as a framework for implementing and evaluating classical and hybrid quantum-classical neural networks. First, we design a variational quantum algorithm for value prediction and a one-dimensional classical convolutional neural network for parallel extraction of protein and drug features. Keeping the classical convolutional neural network fixed, we implement several classical and hybrid quantum-classical models. In the classical approach, three classical prediction modules of different complexities are implemented using three fully connected layers. While in the hybrid approach, five quantum prediction modules are implemented, integrating the variational quantum prediction algorithm with different quantum encodings and quantum circuits of different qubits. The experiments were performed in a simulator and our findings clearly show that hybrid quantum-classical models with lower complexity yield improved accuracy and generalization, demonstrating the quantum advantage. Moreover, the hybrid models speed up the training process in terms of epochs and achieve faster stabilization. The findings suggest that quantum machine learning is a promising direction. Finally, this thesis provides important insights and methodologies that can improve the implementation and evaluation of quantum machine learning approaches, enhancing their practical application in quantum noisy-intermediate-scale devices.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Avramouli, Maria
- Αβραμούλη, Μαρία
Subjects
dc:subject × 20- Κβαντική μηχανική μάθηση
- Ανακάλυψη φαρμάκων
- Υβριδική κβαντο–κλασική μηχανική μάθηση
- Κβαντική υπολογιστική
- Μηχανική μάθηση
- Κβαντικό πλεονέκτημα
- Συγγένεια πρόσδεσης πρωτεΐνης–φαρμάκου
- Quantum machine learning
- Drug discovery
- Hybrid quantum–classical machine learning
- Quantum computing
- Machine learning
- Quantum advantage
- Protein–ligand binding affinity
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58304