AI Reveals Unexpected Source of Antibiotic Candidates in Prion Proteins

Why It Matters

The discovery expands the searchable proteome for new antibiotics, offering a potential pipeline against drug‑resistant infections and illustrating how AI can redefine target spaces in pharmaceutical research.

Key Takeaways

• AI platform APEX 1.1 screened 19.3M fragments from 2,897 prion proteins.
• 1,179 candidate antimicrobial peptides, dubbed “prionins,” were identified.
• 59 of 75 tested peptides inhibited bacteria; 42 showed strong low‑dose activity.
• Two prionins reduced Acinetobacter skin infection in mice, matching polymyxin B.
• Findings expand antibiotic discovery to misfolded proteins previously linked to neurodegeneration.

Pulse Analysis

The global rise of multidrug‑resistant bacteria has forced scientists to look beyond traditional sources for new antibiotics. One promising avenue is the concept of "encrypted peptides"—short functional sequences hidden within larger proteins that can be liberated and repurposed. Prior AI‑driven projects have mined human proteins, microbial genomes, and venoms, but the recent study pushes the frontier into prion and prion‑like proteins, a class historically associated only with fatal neurodegenerative disorders. By leveraging APEX 1.1, researchers could evaluate millions of peptide fragments at scale, turning a once‑overlooked protein family into a fertile hunting ground for antimicrobial agents.

The Penn team’s workflow combined computational prediction with rigorous experimental validation. From an initial pool of 1,179 predicted prionins, 75 were synthesized and screened against 11 bacterial strains, including carbapenem‑resistant Acinetobacter. Over three‑quarters displayed activity, and 42 were potent at micromolar concentrations, a benchmark for therapeutic relevance. Toxicity assays revealed minimal impact on human cells and red blood cells, and two lead candidates performed on par with polymyxin B in a murine skin infection model. These results demonstrate that AI can not only prioritize candidates but also accelerate the transition from in silico hits to in vivo efficacy.

Beyond immediate drug‑development prospects, the findings hint at a deeper biological connection between protein aggregation and innate immunity. While prionins are not naturally secreted during infection, their existence suggests that misfolded proteins may harbor latent defense mechanisms. For biotech firms, the study opens a novel pipeline that could diversify antibiotic portfolios and attract investment focused on AI‑enabled discovery. Future work will need to address manufacturing scalability, regulatory pathways, and the translation of peptide stability from laboratory to clinic, but the proof‑of‑concept establishes prion proteins as a credible, untapped source in the fight against antimicrobial resistance.