Images from this study show correlations between pairs of pharmacokinetic (PK) traits of a single drug. Each drug has its own set of chemical profiles and PK property values. The goal of the figure is to show similarity of distributions between actual pairs of correlations of PK characteristics from in vitro studies and studies generated by researcher models. Credit: Arxiv (2024). doi:10.48550/arxiv.2408.07636
Developing new drugs to treat illnesses has usually been a slow and expensive process. However, a team of researchers at the University of Waterloo uses machine learning to speed up development time.
The Waterloo research team has created Imagand, a generative artificial intelligence model that evaluates existing information about potential drugs and proposes potential properties. Trained and tested with existing drug data, Imagand successfully predicted the key properties of various drugs already independently validated in lab studies, demonstrating the accuracy of AI.
Research entitled “Digrams Smiles-to-Pharmacokinetics Spreading Model with Deep Molecular Understanding” is currently available on ARXIV preprint servers.
Traditionally, bringing successful drug candidates to the market costs between US$2 billion and US$3 billion, and takes over 10 years. Generated AI is poised to transform drug discovery by leveraging large amounts of drug data in diverse regions.
“There is a huge pool of possible chemicals and proteins to investigate when new drugs are developed, making drug discovery extremely expensive because you have to test millions of molecules with thousands of different targets,” says the doctorate. Computer Science Candidates and Research Chief Author. “We’re thinking about ways AI can make it faster and cheaper.”
One of the major challenges in drug development is understanding not only how drugs affect the body when isolated, but also how they interact with other drugs and people’s lifestyles. This information is particularly difficult to collect, as scientific research on drugs usually focuses solely on the given properties of the drug, rather than how they interact with other drugs.
Ultimately, the team hopes that medical researchers can understand how drugs interact with using imaging in the future, eliminating potential new drug candidates with poor side effects and poor interactions.
“For example, this AI-enabled process helps us understand how drugs are toxic, how they affect the heart, or that they can interact negatively with other drugs commonly used to treat diseases,” says Helen Chen, professor of public health and computer science at Waterloo. “This is an example of how AI can help move towards more accurate and personalized care.”
Details: Bing Hu et al, Smiling diffusion models of drug discovery, Deep molecular understanding, Arxiv (2024). doi:10.48550/arxiv.2408.07636
Journal Information: arxiv
Provided by the University of Waterloo
Quote: AI Model predicts drug properties to speed up development acquired on April 3, 2025 from https://phys.org/news/2025-04-ai-drug-properties.html (April 2, 2025)
This document is subject to copyright. Apart from fair transactions for private research or research purposes, there is no part that is reproduced without written permission. Content is provided with information only.
