Ohio State researchers have created a pioneering new generative AI model called DiffSMol, which uses diffusion-based technology to produce molecules with desired properties. Through its application, this unique technology could significantly speed up the drug development process. This creates realistic three-dimensional (3D) structures of small molecules which are used as cellular signalling molecules, making small molecules more promising candidates for new drugs. Those results were released in the journal Nature Machine Intelligence. Specifically, they demonstrate that DiffSMol outperforms state-of-the-art methods to identify the best candidates promising as viable drug candidates.
DiffSMol looks at the geometries of known ligands—molecules that typically interact with a target protein. It then uses this data to generate new 3D molecules that have improved binding properties. This approach speeds the process of drug design while seeking to make pharmaceutical development more cost efficient. DiffSMol has an impressive 61.4% success rate in exhibiting high-quality drug candidates. This performance is a drastic improvement over previous attempts at research, which only succeeded roughly 12% of the time.
Advancements in Drug Development
The creation of DiffSMol represents a new breakthrough for the development of a drug discovery methodology. Because traditional approaches still require significant time and resources. Consequently, the hunt for new drug candidates is a lengthy and costly affair. DiffSMol simplifies this process significantly. Powered by generative AI technology, it takes a fresh look at today’s molecular shapes and generates brand new candidates purpose-built to fit targeted proteins.
Xia Ning Xia is a professor of biomedical informatics and computer science and engineering at The Ohio State University. In her remarks, she underscored the profound potential impact of such technology. “It’s very encouraging for us to find molecules with even better properties than known ligands,” Ning stated.
In practical applications, DiffSMol was further validated on the molecules targeting cyclin-dependent kinase 6 (CDK6) and neprilysin (NEP). CDK6’s fundamental role of regulating a cell’s cycle has important implications for all cancers and cancer therapies. For therapies that seek to slow the progression of Alzheimer’s disease, NEP’s role is critical. DiffSMol’s ability to generate molecules for these high-priority targets highlights the potential of this tool to impact healthcare.
A New Era of Molecular Generation
The process underlying DiffSMol is that the AI model is trained on a dataset of molecular shapes that scientists already understand. Researchers can now prompt the model to create completely new molecules. These new molecules, while sharing similar shapes, will not be present in current chemical databases. Herein lies the uniqueness of this approach, that creates uncharted waters in the drug discovery process.
“By using well-known shapes as a condition, we can train our model to generate novel molecules with similar shapes that don’t exist in previous chemical databases,” Ning explained. This approach equips researchers to customize new drug candidates with greater precision, in the service of identifying and filling specific gaps in therapeutic landscapes.
Despite its successes, DiffSMol remains constrained by only being able to generate new molecules around known ligands. The research team acknowledges this constraint and aims to enhance the model’s capabilities in future iterations, hoping to push the boundaries of what’s achievable in molecular design.
Future Prospects and Challenges
Looking forward, researchers have high hopes for generative AI improving the drug design process even further. This field is expanding at an exponential rate. As for the future, these innovations will continue building upon each other, enhancing the already impressive capabilities of models like DiffSMol. “Nowadays, people are applying these advanced models to molecule generation, to chemistry, to nearly all science areas,” Ning noted, highlighting the widespread implications of this technology.
In addition, Ning was optimistic that this non-residential building boom could be sustainable. “This area grows really fast and I don’t see it slowing down anytime soon.” In the meantime, as the research community digs deeper into generative AI’s capabilities, the prospects for discovering novel drug candidates seems highly encouraging.
