Researchers at the University of Pennsylvania used artificial intelligence to search for short peptides that could serve as starting points for new antibiotics. The team employed a machine‑learning system called APEX to analyze 19.3 million short peptide fragments associated with 2,897 prion and prion‑like proteins sourced from humans, mice, cows, naked mole‑rats and other animals.
The AI screen identified 1,179 antimicrobial peptide candidates, which the authors referred to as “prionins.” From those, the researchers selected the 75 most promising peptides for laboratory testing against 11 different bacterial species, including drug‑resistant strains.
In vitro assays showed that at least 59 of the 75 peptides inhibited at least one bacterial pathogen, and 42 displayed strong activity at low concentrations. The peptides appeared to act primarily by disrupting bacterial membranes, a mechanism similar to many known peptide antibiotics. The team published their results in the journal Nature Microbiology.
Why the source matters and what risks arise
Prions are protein‑based infectious agents that can convert normally folded proteins of the same species into misfolded, self‑propagating forms, producing additional prions. Although prions do not contain genetic material and are not living organisms in the conventional sense, they can cause severe, often fatal diseases in hosts, such as Creutzfeldt–Jakob disease. Prions are also resistant to standard sterilization methods (for example, boiling), which makes infections difficult to eliminate and prion diseases essentially untreatable with conventional approaches.
Because the candidate peptides originate from prion and prion‑like proteins, their source raises biosafety considerations: material derived from prion proteins is linked to infectious structures that require special handling. The researchers note that while the antimicrobial activity is promising, further study and development must proceed with caution to address potential biological risks associated with prion‑derived materials.
Implications for drug discovery
The study suggests that peptides found in human proteins, extinct organisms, or other unconventional sources may contain useful antimicrobial motifs, and that AI enables large‑scale screening of millions of fragments to find them. This capability represents a meaningful advance for antibiotic discovery, provided that work on prion‑associated peptides accounts for the additional biosafety and handling challenges their origin implies.
The work was reported in part by IFLScience and published in Nature Microbiology.
