Researchers at the University of Pennsylvania, led by César de la Fuente, utilized an updated version of their deep-learning AI tool called APEX (Antimicrobial Peptide Explorer) to analyze the proteomes of 233 species of Archaea.
Archaea are extremophiles that thrive in harsh conditions and have evolved unique biochemical mechanisms over billions of years, making them a largely untapped source for novel compounds. The AI screened 18,677 non-redundant protein sequences from these organisms, predicting 12,623 peptides (short protein fragments) with potential antimicrobial properties. These compounds, dubbed “archaeasins,” were validated in lab tests against various bacteria, including drug-resistant strains like Escherichia coli and Staphylococcus aureus.
Unlike many traditional antibiotics that target bacterial cell walls or outer membranes, archaeasins seem to disrupt the bacteria’s internal electrochemical gradients—essentially interfering with the proton motive force that powers cellular processes, leading to the cell’s shutdown from within.
The study, published in Nature Microbiology on August 12, 2025, highlights how AI can accelerate drug discovery by mining ancient microbial genomes for solutions to modern antibiotic resistance. The team plans to further refine APEX for better structural predictions and safety assessments of these archaeasins.
This work builds on earlier efforts by the same lab, which used similar AI methods to identify antibiotics from other sources like the human microbiome. While promising, these are still in early stages, and clinical applications would require extensive testing.
AI just found 12,000 possible new antibiotics hiding in ancient microbes.
— Brian Roemmele (@BrianRoemmele) August 14, 2025
Antibiotic resistance is one of humanity’s biggest health threats, killing an estimated 1.27 million people a year – and scientists warn it could get far worse. But a new study suggests the next lifesaving… pic.twitter.com/CBz3UonfAY