AI turns up antibiotic peptides hidden in archaea, life’s oddest branch

Where they looked

Archaea look like bacteria under the microscope but are a separate branch of life. Many live in hot springs, salt lakes and the guts of animals, competing with bacteria for space. The team guessed that competition might leave traces in their proteins.

They used a deep-learning model, an updated version of a tool called APEX, to scan the proteomes of hundreds of archaea for short stretches likely to kill bacteria. The model flagged more than 12,000 candidates, which the team named archaeasins.

What held up

The lab made 80 of them. About nine in ten killed at least one bacterial species in a dish, including drug-resistant strains. Several worked by disrupting the electrical charge across the bacterial membrane, a slightly different route from the classic hole-punching of many antimicrobial peptides.

In mice with skin abscesses caused by Acinetobacter baumannii, the best candidate, archaeasin-73, reduced bacterial counts about as much as polymyxin B, an old last-resort antibiotic.

The usual caveat

Antimicrobial peptides often look strong in a dish and then stumble on toxicity, stability in blood and cost. None of the archaeasins has gone near a human trial. The paper’s main contribution is a new hunting ground and evidence that the AI screen finds real activity much more often than chance.

Source Torres M. D. T. et al., Nature Microbiology, August 2025

Amino Weekly reports research. It is not medical advice and never gives doses.