Deep learning reveals antimicrobial peptides within prions

Introduction

Prions are best known for their role in rare, fatal neurodegenerative diseases, where misfolding and aggregation can have devastating effects. But our latest study suggests that prion and prion-like proteins may also contain something very different: short molecular fragments capable of killing bacteria, including drug-resistant pathogens.

In a paper published in Nature Microbiology, our team used a deep-learning platform called APEX 1.1 to search for antimicrobial activity hidden inside prion-related proteins. The scale of the search was enormous: APEX 1.1 scanned 19.3 million short protein fragments derived from 2,897 prion and prion-like proteins. From this search, the model identified 1,179 candidate antimicrobial peptides, which we named prionins.

Why this was unexpected?

Prions are not an obvious place to look for new antibiotics. They are most often associated with disease, not immunity or antimicrobial defense.

Still, previous studies had offered hints that amyloid-associated sequences, including amyloid-β and the cellular prion protein, can show antimicrobial or host-protective activity. Our study asked whether those scattered observations pointed to something broader.

From prediction to experiment

To test the AI predictions, we synthesized 75 prionins and screened them against clinically relevant bacterial pathogens, including multidrug-resistant strains.

The results were striking: 59 of the 75 peptides inhibited at least one pathogen, and 42 showed potent activity at concentrations of 16 micromolar or lower against at least one pathogen.

How prionins kill bacteria

We then asked how these molecules were working. Many active prionins damaged bacterial membranes, a common mechanism used by antimicrobial peptides.

Just as importantly, several candidates showed early signs of selectivity. Hemolysis was rare, and 16 active peptides showed neither measurable hemolysis nor cytotoxicity at the highest concentrations tested.

Testing the strongest candidates in mice

Two of the strongest candidates were tested in a mouse skin-infection model caused by Acinetobacter baumannii, a difficult-to-treat pathogen.

A single topical dose of each peptide significantly reduced bacterial burden, with effects comparable to polymyxin B in the model tested. The researchers observed no treatment-associated weight loss.

Conclusion

The findings build on our lab’s broader effort to mine the biological world for encrypted peptides: short, hidden sequences inside larger proteins that can have biological functions when isolated.

The study raises intriguing questions at the intersection of neurodegeneration and innate immunity. However, it does not show that prionins are naturally released during infection or that they function physiologically as immune molecules.

It also does not change the established role of misfolded prions in devastating neurodegenerative disease.

Instead, the work identifies prion and prion-like proteins as a previously overlooked source space for antibiotic discovery.

For more information on this study, please refer to the full paper published in Nature Microbioloy: https://www.nature.com/articles/s42256-026-01237-5

For more information, please contact:

Machine Biology Group

University of Pennsylvania

Authors:

Marcelo D. T. Torres, Fangping Wan, & Cesar de la Fuente-Nunez

Published:

June 19, 2026

About Machine Biology Group:

The mission statement of the Machine Biology Group at the University of Pennsylvania is to use the power of machines to accelerate discoveries in biology and medicine.

Posted in Uncategorized.