Model generates novel antibiotic structures
Researchers at the University of Pennsylvania have introduced a groundbreaking AI tool called AMP-Diffusion. This innovative technology is designed to create antimicrobial peptides (AMPs) with potential bacteria-killing properties.
The design of AMP-Diffusion allows for a more streamlined and focused approach to creating potential antimicrobial drugs. It generates candidate AMPs more efficiently, thanks to its use of a pre-trained protein language model.
Pranam Chatterjee, assistant professor in bioengineering and computer science, and the senior co-author of the paper introducing AMP-Diffusion, compares the approach to starting with a fluent speaker, rather than teaching the model the basics. Chatterjee's team started with a model that already has a rich understanding of protein structures, rather than teaching the model basic biology.
AMP-Diffusion doesn't need to relearn basic biology because it uses ESM-2, a widely-used protein language model, which already has a rich "mental map" of how real proteins fit together. This understanding enables AMP-Diffusion to create candidate AMPs that are more likely to follow the intricate patterns that make peptides effective.
To refine sequences of amino acids, AMP-Diffusion uses generative AI algorithms similar to those used for image and text generation. However, instead of "denoising" pixels, AMP-Diffusion refines sequences of amino acids. AMP-Diffusion's outputs are more likely to follow the intricate patterns that make peptides effective, due to its use of ESM-2.
The focus of Chatterjee's team is on designing effective antimicrobial peptides, rather than teaching the model basic biology. AMP-Diffusion is given a coach that keeps it grounded in biological reality. This means AMP-Diffusion can generate candidate AMPs faster and its outputs are more likely to follow the intricate patterns that make peptides effective.
The results of the research, published in the journal Cell Biomaterials, are promising. The most potent AMPs created by AMP-Diffusion performed as well as FDA-approved drugs in animal models, without detectable adverse effects.
The two senior co-authors of the paper, apart from Chatterjee, are José R. Torero and Alan J. Wolf, both affiliated with the University of Pennsylvania's Perelman School of Medicine. César de la Fuente, Presidential Associate Professor in multiple fields, is the other senior co-author of the paper introducing AMP-Diffusion.
This development could pave the way for faster and more efficient creation of antimicrobial drugs, potentially revolutionising the field of medicine.
