AI discovers hidden antibiotic candidates inside prion proteins linked to brain diseases

AI Uncovers Antibiotic Potential in Prion Proteins

Researchers at the University of Pennsylvania have identified a startling new frontier in antibiotic discovery by using AI to search for hidden antimicrobial compounds within prion proteins, the misfolded brain proteins typically associated with rare and fatal neurodegenerative diseases. The findings, published in Nature Microbiology, reveal that prion and prion-like proteins harbor short peptides dubbed "prionins" that can kill bacteria, opening an unexpected avenue for combating drug-resistant infections at a time when treatment options are increasingly limited 1

1

2

2

.

Deep-Learning Platform Scans Millions of Peptide Fragments

The research team deployed an AI deep-learning platform called APEX 1.1 to systematically scan 19.3 million short peptide fragments extracted from 2,897 prion and prion-like proteins. This computational approach allowed researchers to predict the antibiotic activity of specific amino acid sequences at a scale impossible through traditional laboratory methods. The platform identified 1,179 candidate antimicrobial peptides, establishing a new class of antibiotic candidates that the team named prionins 1

1

. "This work changes where we think antibiotics might be hiding," said César de la Fuente, PhD, Presidential Associate Professor and director of the Machine Biology Group at the University of Pennsylvania Perelman School of Medicine. "Prions have long been seen almost entirely through the lens of disease, but AI let us ask a different question: whether these proteins also encode useful molecular fragments. The answer appears to be yes" 2

2

.

Lab Testing Validates Antimicrobial Peptides Against Drug-Resistant Strains

From the pool of AI discovers hidden antibiotic candidates, researchers selected 75 of the most promising peptides for experimental validation against 11 different bacterial pathogens, including drug-resistant strains. The results proved compelling: 59 peptides inhibited at least one bacterial pathogen, while 42 demonstrated strong activity at low concentrations. Additional experiments revealed that many active prionins work by disrupting bacterial membranes, a well-established mechanism employed by antimicrobial peptides. Importantly, toxicity testing showed limited harm, with 16 active peptides displaying no measurable damage to red blood cells or human cells at the highest concentrations tested 1

1

.

Mouse Infection Models Demonstrate Real-World Efficacy

To move beyond laboratory predictions, the team tested two of the most promising peptides—one derived from a fungus and one from a roundworm—in mouse infection models. The peptides successfully reduced bacterial levels in a standard skin infection caused by Acinetobacter baumannii, a notoriously difficult-to-treat pathogen. Their effectiveness proved comparable to polymyxin B, an established antibiotic, and researchers observed no treatment-related weight loss in the animals. "The AI search gave us a short list of candidates, but the important point is that many of those molecules worked in the lab, and two worked in an animal infection model," said Marcelo D. T. Torres, co-first author of the study. "That is what makes this a discovery platform, not just a prediction exercise" 2

2

.

Encrypted Peptides Connect Neurodegeneration and Innate Immunity

The discovery builds on the de la Fuente Lab's broader initiative to mine the biological world for encrypted peptides—short, hidden sequences within larger proteins that can exhibit biological functions when isolated. Previous research from the group has explored human proteins, extinct organisms, archaea, microbiomes, and venoms, but the prion study ventures into one of biology's most unexpected protein classes 1

1

. The work raises intriguing questions about potential links between protein aggregation and host defense mechanisms, though it does not suggest that prionins are naturally released during infection or that prion proteins normally function as antibiotics in the body. The harmful role of misfolded prions in neurodegenerative diseases remains unchanged, but these proteins may represent a rich and previously overlooked reservoir for antibiotic development. Earlier studies had hinted that fragments from proteins like amyloid-beta, involved in Alzheimer's disease, and the cellular prion protein could fight microbes, but no one had systematically searched prion and prion-like proteins at scale until now 2

2

.

Why This Matters for Drug Development

The research arrives at a critical juncture as drug-resistant infections continue to narrow treatment options globally. By demonstrating that AI can identify functional antimicrobial sequences hidden within proteins traditionally viewed only through the lens of disease, the study expands the search space for new antibiotics dramatically. The approach offers a systematic method to explore biological sources that conventional drug discovery might never consider. "For a long time, drug discovery has been limited not only by what we can test, but by where we choose to look," de la Fuente noted. "AI is changing that. It gives us a way to search the hidden layers" of biology 1

1

. As researchers continue to validate prionins and explore their therapeutic potential, the intersection of AI, neuroscience, and infectious disease may yield additional breakthroughs in addressing one of medicine's most pressing challenges.