TL;DR
César de la Fuente at the University of Pennsylvania is training AI models to scan the genomes of extinct species — from woolly mammoths to Neanderthals — for peptides with antibiotic properties. His team has built a library of over a million genetic recipes and successfully treated drug-resistant infections in mice using AI-designed synthetic peptides.
Mining extinct DNA for new drugs
Antimicrobial resistance now causes more than 4 million deaths per year, with projections suggesting that figure could exceed 8 million by 2050. The antibiotic discovery pipeline remains thin, impeded by high development costs and poor returns on investment. Many companies that attempted antibiotic development have folded.
De la Fuente’s approach turns biology into an information problem. His team trains AI models to recognise sequences of amino acids that encode antimicrobial peptides (AMPs) — small molecules that the immune system already uses as a first line of defence against infections. Unlike conventional antibiotics, which typically have a single mechanism for killing bacteria, AMPs often attack through multiple pathways simultaneously, making it harder for bacteria to evolve resistance.
The team has found candidates in unexpected places: the genetic code of archaea (ancient single-celled organisms), the venom of snakes and spiders, and the DNA of extinct species. A project called “molecular de-extinction” has produced compounds named mammuthusin-2 (from woolly mammoth DNA), mylodonin-2 (from the giant sloth), and hydrodamin-1 (from an ancient sea cow).
From prediction to generation
The field has shifted from using predictive AI models that screen known candidates to generative approaches that design new molecules from scratch. Last year, de la Fuente’s team used generative AI to create synthetic peptides and tested two against a drug-resistant strain of Acinetobacter baumannii, classified by the WHO as a “critical priority” pathogen. Both successfully treated the infection in mice.
His team is now developing a multimodal model called ApexOracle, designed to analyse a pathogen, identify its genetic weaknesses, match it to effective peptides, and predict clinical performance.
Looking forward
While these peptides have not yet become usable drugs — with dosage, delivery, and targeting still to be resolved — the work demonstrates AI’s potential to accelerate the discovery phase of drug development. For the UK’s life sciences sector, which faces similar antimicrobial resistance pressures, these approaches represent a growing area of research collaboration and investment.