There’s now a ChatGPT for molecules and it hopes to accelerate drug discovery

Researchers at the University of Bonn have developed an AI tool to design dual-target drug molecules, potentially transforming pharmaceutical research.

Scientists at the University of Bonn have developed a chemical language model published this week in Cell Reports Physical Science, a type of AI trained to predict new drug compounds with dual-target capabilities. The AI, described as a “ChatGPT for molecules,” generates chemical structures that can bind to two different proteins simultaneously. This could pave the way for treatments that are more effective for complex diseases such as cancer.

The AI was trained on more than 70,000 pairs of molecular structures. One string described a compound targeting one protein, while the other represented a compound that could influence two. After this training, the model successfully predicted compounds with dual-target activities—an especially valuable property in pharmaceutical research.

Multi-target drugs: A breakthrough in medicine?

Dual-target drugs, also known as multi-target compounds, have significant potential for improving therapies. Prof. Dr. Jürgen Bajorath, head of the research, highlighted the importance of such compounds: “In pharmaceutical research, these types of active compounds are highly desirable due to their polypharmacology.”

Compounds that target multiple biological pathways are often more effective, particularly in fields like oncology, where disrupting multiple processes simultaneously can improve treatment outcomes. Traditional approaches rely on administering multiple drugs, which can lead to adverse drug interactions or metabolic inconsistencies. However, by designing single compounds that act on two proteins, researchers can reduce such risks.

Beyond cancer treatment, this approach could be applied to other complex diseases, such as neurodegenerative disorders. In Alzheimer’s disease, for example, targeting both beta-amyloid plaques and tau proteins could provide a more comprehensive treatment, potentially slowing disease progression more effectively than single-target therapies.

How the AI works

The chemical language model utilises a system known as SMILES (Simplified Molecular Input Line Entry System), a notation used to describe chemical structures. Feeding the AI pairs of these SMILES strings allowed it to learn how dual-target molecules differ from those acting on a single target. After fine-tuning, the AI was able to suggest entirely new compounds with dual-target potential—some of which have already been validated as effective.

“We trained our model to suggest molecules that act not only against one protein but also against another,” said Sanjana Srinivasan, a researcher from Bajorath’s group.

The AI’s ability to work with relatively small data sets is particularly noteworthy, as traditional machine learning systems often require enormous datasets. This makes it feasible to apply chemical language models in research areas where data may be limited, such as rare diseases.

Unconventional solutions for future drug design

Although the AI tool may not immediately generate compounds superior to current drugs, its real strength lies in its ability to offer unexpected solutions. “The AI often suggests chemical structures that most chemists would not even think of right away,” Bajorath explained. This capacity to think “out of the box” could open up new avenues for drug design that human researchers might overlook.

Moreover, these unconventional structures could inspire new research directions. For instance, the AI’s suggestions might lead to the development of compounds that target previously unconsidered pathways or molecular interactions, broadening the scope of pharmaceutical innovation.

Expanding beyond pharmaceuticals

While the AI’s primary application is in drug discovery, it has broader potential. The underlying principles of the chemical language model could be applied to other industries, including materials science and environmental chemistry. Researchers foresee the possibility of designing new materials with specific properties, or even creating environmentally friendly chemicals for use in consumer products.

The future of AI in chemistry is still unfolding, but tools like this chemical language model are likely to play a critical role in scientific innovation across multiple sectors. As AI systems become more sophisticated, their ability to generate novel solutions and drive breakthroughs will only continue to expand.