Predicting Drug-Target Interaction with Machine Learning

Thumbnail Image

Files

Särkiö_Liisa_2025.pdf (709.52 KB)

URL

Journal Title

Journal ISSN

Volume Title

Sähkötekniikan korkeakoulu | Bachelor's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2025-05-30

Department

Major/Subject

Bioinformaatioteknologia

Mcode

BIO

Degree programme

Sähkötekniikan kandidaattiohjelma

Language

en

Pages

22+6

Series

Abstract

Bringing a new drug into the market is a long and tedious process of ensuring the safety and efficiency of that drug. During this process, it is crucial to test how the substance reacts with both its intended target and other structures in the human body to avoid unwanted and dangerous side effects. However, manually examining the interactions between the drug compound and e.g. the proteins in the body in a laboratory is extremely costly and time-consuming. Thus, researchers have long developed different kinds of computational methods to determine the existence and nature of these interactions. In addition to computational methods, during the last decades machine learning methods have been developed, that can further accelerate the drug development process and mimic the decision making of the human brain. This bachelor’s thesis inspects how machine learning can be used in predicting drug-target interaction (DTI). The thesis focuses on three machine learning methods: random forest, kernel methods, and deep learning. These methods are chosen for closer examination, because they have often been used for DTI prediction. This study is executed as a literature review, and it presents and compares different research examples where the aforementioned machine learning methods have been used for predicting drug-target interaction. By examining research examples, it is noticed that in most cases machine learning succeeds in predicting interaction with relatively good accuracy. Random forest, kernel methods, and deep learning all have their strengths and weaknesses, even though all of them can be used to make satisfactory predictions on DTI. The differences in the method have to do with e.g. the readability of the model, computational costs, and the data and setup that is most suitable for each model. Based on the literature review conducted, deep learning seems to have the most potential, so it is justifiable to assume that it will play an even bigger role in the future. The thesis makes the conclusion that machine learning is a significant tool, from which the pharmaceutical industry can greatly benefit from, if its use is regulated enough. Challenges were faced especially in scenarios where the targets of the predicted drug were previously relatively unknown. This proves that, at least as of now, machine learning cannot completely replace traditional experimental methods of drug discovery. However, in tandem with them, machine learning can provide solutions to problems that have long puzzled the medical community. The next important step is to further incorporate machine learning from research into the actual drug discovery process.

Description

Supervisor

Turunen, Markus

Thesis advisor

Rousu, Juho

Keywords

pharmaceutical industry, proteins, random forest, kernel methods, deep learning

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202506034195

Collections

[kand] Sähkötekniikan korkeakoulu / ELEC