Penn Engineering reveals AI model that designs new antibiotics.
Credit: Abhiram Juvvadi
Researchers at Penn’s School of Engineering and Applied Science introduced a new generative artificial intelligence tool to create antibiotics.
In a recent paper — titled “Generative latent diffusion language modeling yields anti-infective synthetic peptides” — the researchers presented AMP-Diffusion, an AI platform used to synthesize antimicrobial peptides. The team of researchers included Perelman School of Medicine research associate Marcelo Torres, postdoctoral researcher Fangping Wan, assistant Bioengineering professor Pranam Chatterjee, and Psychiatry professor César de la Fuente-Nunez.
“Nature’s dataset is finite; with AI, we can design antibiotics evolution never tried,” de la Fuente told Penn Engineering Today.
Chatterjee explained that the team leveraged the “same AI algorithms that generate images,” but altered them to “design potent new molecules.”
To develop the tool, de la Fuente’s lab joined forces with Chatterjee’s, combining de la Fuente’s lab’s experience with detecting molecules with antimicrobial properties with Chatterjee’s lab’s experience designing peptides for treating diseases.
“It seemed like a natural fit,” Chatterjee said. “Our lab knows how to design new molecules using AI, and the de la Fuente Lab knows how to identify strong antibiotic candidates using AI.”
AMP-Diffusion works by using latent diffusion modeling to create AMPs, which can kill harmful bacteria. The paper explained that the technology “enables the rapid discovery of antibiotic candidates by systematically exploring sequence space.”
After generating about 50,000 amino acid sequences, the researchers needed to narrow down their list of candidate drugs — using AI to do so. The team used APEX 1.1, a deep learning model developed by de la Fuente’s lab, to select 46 peptides for the experiment. After synthesizing the peptides, the lab tested them using human cells and mice.
Testing two lead candidates, AMP-diff2-16 and 43, for in vivo efficacy, the researchers infected mice with A. baumannii cells. They administered a single dose of each peptide directly to the infected area, and found the results encouraging.
Sign up for our newsletter
Get our newsletter, DP Daybreak, delivered to your inbox every weekday morning.
“After two days, both peptides markedly reduced the bacterial load by 1.5 orders of magnitude,” the paper said. “After four days, bacterial counts had decreased by 2-2.5 orders of magnitude, comparable to reductions achieved by widely used antibiotics.”
According to the paper, 35 of the 46 peptides tested were effective against at least one of the 11 bacterial pathogens tested. The paper added that no weight changes, skin damage, or other adverse effects were observed in treated mice.
“It’s exciting to see that our AI-generated molecules actually worked,” Chatterjee said. “This shows that generative AI can help combat antibiotic resistance.”
The Daily Pennsylvanian is an independent, student-run newspaper. Please consider making a donation to support the coverage that shapes the University. Your generosity ensures a future of strong journalism at Penn.