Using AI to combat the antibiotic resistance crisis
“Antibiotic resistant infections are one of the greatest existential threats facing humanity,” said Dr. César de la Fuente, Presidential Associate Professor at the University of Pennsylvania.
In his keynote presentation at Technology Networks’ Innovations in Disease Modeling 2025 event, Dr. de la Fuente outlined how his lab is using artificial intelligence (AI) to reimagine antibiotic discovery – by turning to the human proteome, ancient genomes and even extinct organisms.
Each year, bacterial infections kill five million people globally. By 2050, that number could double, potentially surpassing deaths from cancer.
Traditional antibiotic discovery is slow and costly, often relying on trial and error. Dr. de la Fuente’s response? “Why not conceptualize biology as an information source—a bunch of code—that can be explored with the right algorithms to try to find new hidden molecules?”
Peptides: Nature’s programmable nanomachines
Peptides, short chains of amino acids, are at the heart of this work. These molecules “are the simplest nanomachines that can do functional activities in biology,” and are well-suited for AI analysis because of their diversity and scalability. Dr. de la Fuente’s lab uses AI models to mine the vast “sequence space” of peptides for molecules with antimicrobial, anticancer or immunomodulatory properties.
From human proteome to encrypted antibiotics
Using AI scoring functions inspired by image and speech recognition models, the team began scanning the entire human proteome – over 42,000 proteins – for encrypted antimicrobial peptides. Remarkably, “our algorithm… was capable of sampling through every protein in our body in just about one hour.” These predictions led to the discovery of thousands of new antibiotic candidates, over 60% of which were experimentally validated in the lab.
Molecular de-extinction: Mining ancient genomes
Inspired by these results, the team expanded their search to extinct relatives like Neanderthals and Denisovans. “We came up with a new framework for identifying molecules… which we call molecular de-extinction,” Dr. de la Fuente explained. The process resurrects ancient peptide sequences using AI and synthetic chemistry, testing their efficacy against modern pathogens in mouse models.
One such molecule, “Neanderthalin,” showed promising anti-infective activity, comparable to last-resort antibiotics like polymyxin B.
APEX and generative AI: Scaling drug discovery to prehistoric depths
To scale this approach across all extinct life, the team developed APEX, a custom-trained AI model capable of predicting peptide function from sequence alone. The result? Ancient elephants, sea cows and even magnolia trees yielded viable antibiotic candidates.
His lab recently introduced APEX-GO, a generative AI that designs improved peptide analogs, achieving an 85% hit rate in lab tests and a 72% success rate in improving antibiotic potency. “This just opens new avenues for optimizing molecules,” said Dr. de la Fuente.
Peptides that do more: Designing multimodal therapeutics
The team also created APEX-DUO, a multimodal AI system that designs peptides with more than one function, such as penetrating human cells and killing intracellular bacteria. “In the future, you can think of designing new medicines that can do two things,” he added, such as combining antibacterial and anti-inflammatory actions.
Bioethics, patents and the future of AI in drug discovery
With power comes responsibility. “We started consulting with bioethicists to make sure that we innovated, but that we did so responsibly,” said Dr. de la Fuente. His lab refrains from synthesizing molecules that resemble biotoxins or bio-weapons.
The project also raises new legal questions: can ancient molecules – once natural but now extinct – be patented? “This is actually opening up a new area of patent law,” he noted.
Looking ahead: AI at the Center of Scientific Innovation
Dr. de la Fuente estimates that his team’s work has accelerated the field of antibiotic discovery by a staggering “one million years of research time”, or “the equivalent of around 100,000 PhD students working for 6 years each.”
His lab’s AI-first approach is expanding beyond antibiotics into neuroscience, cancer, agriculture and food science.
Reflecting on initial skepticism, he shared: “People thought we were on the lunatic fringe… but sometimes, if you really believe in something… you can really show to the world that something that seems impossible might be possible.”
This content includes text that has been generated with the assistance of AI. Technology Networks’ AI policy can be found here.
