University of Pennsylvania researchers used deep learning (APEX 1.1) to scan thousands of prion and prion-like proteins and identify more than 1,000 candidate antimicrobial peptides termed “prionins.” In synthesis and functional tests, 59 peptides inhibited bacterial pathogens. The team reported that two prionins reduced Acinetobacter baumannii burden in mice. The study was published in Nature Microbiology, highlighting an unexpected antimicrobial potential in aggregation-prone proteins long associated mainly with neurodegenerative disease. The findings also support a broader drug discovery concept: AI-guided mining of “cryptic” fragments within protein families may yield antibiotic leads amid rising resistance.