Researchers at the School of Engineering and Applied Science have developed an artificial intelligence-powered model that examines and improves antibiotic candidates.
The model, called ApexGO, builds upon a model which was released two years ago and tells researchers whether a given peptide is likely to have antimicrobial properties. The latest iteration goes beyond analyzing a set of molecules and suggests improvements to make more effective antibiotics.
Presidential Associate Professor César de la Fuente characterized ApexGO — which stands for APEX generative optimization — as “a way of improving lousy antimicrobials” in an interview with The Daily Pennsylvanian.
“I think 85% of everything that it creates that we synthesized was capable of killing bacteria in real-world experiments,” de la Fuente said, adding that “72% of the variants that it created were better than the template molecule at killing bacteria.”
Research assistant professor of psychiatry Marcelo Torres wrote in a statement to the DP that the model “learns patterns from biological sequences and uses this information to prioritize candidates that are more likely to have antibacterial activity.”
“Instead of testing molecules randomly, the system balances what it predicts to be promising with areas where the model is still uncertain,” Torres wrote. “This helps the platform identify candidates that are both potentially active and informative for improving the search.”
2025 Engineering School Ph.D. graduate Natalie Maus explained that, because there are so many possible peptides that could have antibiotic properties, “you could do wet lab experiments all day every day for a year, and you would be testing such a small fraction of the space of possibilities.”
She explained that the ApexGO model allows researchers to focus on a smaller subgroup of peptides which can be more easily tested and verified.
RELATED:
CHOP hosts former Penn Med professor at immunology symposium
Penn Engineering changes email domain to ‘unify’ school identity
Before graduating, Maus worked on ApexGO with assistant professor of computer and information sciences Jacob Gardner — the paper’s other senior co-author.
Maus added that when the researchers tested peptides generated by ApexGO in living mice, they “achieved even stronger inhibitory activity than some of these FDA-approved last resort antibiotics,” which are “reserved for special cases of very highly antibiotic-resistant pathogens.”
De la Fuente emphasized the importance of developing new antibiotics, characterizing “antimicrobial-resistant infections” as “the greatest existential threat to humanity” that “nobody knows about.”
According to de la Fuente, these infections “are associated with five million deaths per year around the globe today” — a number which he said could rise to 10 million by 2050 if scientists don’t “come up with new antimicrobial strategies.”
He added that tools like APEX and ApexGO are “very helpful” to the development of new antibiotics that “eventually can save humanity.”
Similarly, Maus described ApexGO as “an exciting step towards this future.”
“The broader implication is that ApexGO could help make the early stages of antibiotic discovery faster, more systematic, and more data-driven.” Torres wrote. “More broadly, these methods could be applied to other areas of peptide and protein engineering where the search space is too large to explore experimentally by trial and error alone.”
Collaboration is also essential to the efficient development of necessary technologies, according to Maus.
“Research like this that’s open and shared with the broader scientific community is important so that we can get as many people on board and help tackle this really hard problem,” she said.
