Chemistry

Chemistry ChatGPT for new drugs: Researchers train AI to predict potential active ingredients

Three-dimensional structures of two target proteins, histone deacetylase 6 (blue) and tyrosine protein kinase JAK2 (red), and selective inhibitors of each enzyme. The central dual inhibitor is active against both targets. Prediction of compounds with predefined dual target activities is the task of chemical language models. Credit: Sanjana Srinivasan & Jürgen Bajorath

Researchers at the University of Bonn have trained an AI process to predict potential active ingredients with special properties, resulting in a chemical language model that is a kind of molecular ChatGPT. After a training phase, the AI ​​was able to accurately reproduce the chemical structures of compounds with known dual-target activity, which could make them particularly effective drugs. The study is now published in Cell Reports Physical Science.

If you want to make your grandma happy with a poem on her 90th birthday, you don’t need to be a poet these days. Just type a short prompt in ChatGPT and the AI ​​will spit out a long list of rhyming words within seconds. along with the birthday girl’s name. You can also compose a sonnet to go with it if you wish.

Researchers at the University of Bonn implemented a similar model, known as the chemical language model, in their study. However, this does not create rhyme. Instead, AI displays the structural formulas of compounds that may have particularly desirable properties that can bind to two different target proteins. In biology, this means that, for example, two enzymes can be inhibited at the same time.

I want an active ingredient that has double effects.

“In pharmaceutical research, these types of active compounds are highly desirable due to their polypharmacality,” explains Professor Jürgen Bajoras. The computational chemistry expert leads the AI ​​in Life Sciences at the Lamar Institute for Machine Learning and Artificial Intelligence and the Life Sciences Informatics program at Unibon’s b-it (Bon Aachen International Center for Information Technology). “Compounds with desirable multitarget activity are often particularly effective, such as in the fight against cancer, because they affect several intracellular processes and signaling pathways simultaneously.”

In principle, this effect can also be achieved by combining different drugs. However, there is a risk of undesirable drug interactions, and different compounds often break down at different rates in the body, making it difficult to administer them together.

Finding molecules that specifically affect the effects of a single target protein is not an easy task. Designing compounds with predefined dual effects is even more complex. Chemical language models may be useful here in the future. ChatGPT is trained on billions of pages of written text and learns to construct sentences itself.

Chemical language models work similarly, but have relatively little data available for training. However, as a rule, texts such as SMILES strings are also provided, which represent organic molecules and their structures as a series of letters and symbols.

“We have now trained a chemical language model using pairs of strings,” says Sanjana Srinivasan of the Bajoras research group. “One of the strings described a molecule that was known to act only on one target protein, and the other described a molecule that was known to affect this protein as well as a second target protein. It represented a compound that gives.

AI learns chemical bonds

The model was given over 70,000 of these pairs. This allowed us to acquire tacit knowledge about how ordinary active compounds differ from compounds with dual effects.

“We then gave them a compound against the target protein, and this suggested a molecule that would not only act on this protein, but also on other proteins,” Bajoras explains.

Dual-effect training compounds often target similar proteins and therefore perform similar functions in the body. However, pharmaceutical research is also looking for active ingredients that affect completely different types of enzymes and receptors.

To prepare the AI ​​for this task, a general learning phase was followed by some fine-tuning. The researchers used dozens of special training pairs to teach the algorithm which different classes of proteins the proposed compound should target. This is a bit like telling ChatGPT to create limericks instead of sonnets this time.

After some fine-tuning, the model actually spits out molecules that have already been shown to act against the desired combination of target proteins. “This shows the process is working,” Bajoras said. However, in his opinion, the strength of this approach is that new compounds that exceed the efficacy of available drugs will not be found anytime soon.

“What’s more interesting from my perspective is that AI often suggests chemical structures that most chemists wouldn’t immediately think of,” he explains. “To a certain extent, it generates ideas ‘outside the box’ and comes up with unique solutions that lead to new design hypotheses and approaches.”

More information: Generation of dual-target compounds using transformer chemical language models, Cell Reports Physical Science (2024). DOI: 10.1016/j.xcrp.2024.102255. www.cell.com/cell-reports-phys … 2666-3864(24)00560-5

Provided by University of Bonn

Citation: ‘Chemistry ChatGPT’ for new drugs: Researchers train AI to predict potential active ingredients (October 23, 2024) https://phys.org/news/2024-10-chemical-chatgpt Retrieved October 23, 2024 from -medications- ai-possibility.html

This document is subject to copyright. No part may be reproduced without written permission, except in fair dealing for personal study or research purposes. Content is provided for informational purposes only.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button