Antimicrobial resistance has escalated into a global health crisis, with infections from drug-resistant bacteria, fungi, and viruses now associated with more than 4 million deaths per year, and a recent analysis predicts that number could surge past 8 million by 2050. Researchers warn of a possible “post antibiotic” era in which once treatable infections from common bacteria such as Escherichia coli and Staphylococcus aureus could become routinely fatal, while the antibiotic discovery pipeline remains thin and economically unattractive. Against this backdrop, bioengineer and computational biologist César de la Fuente is applying artificial intelligence to expand the search for new antimicrobials far beyond traditional methods.
De la Fuente’s group at the University of Pennsylvania trains artificial intelligence models to scan genetic code for antimicrobial peptides, or AMPs, short chains of up to 50 amino acids that form a key part of the immune system’s first line of defense. His team has identified promising peptide candidates in unexpected sources, including ancient single celled archaea, the venoms of snakes, wasps, and spiders, and a “molecular de extinction” project that mines published genetic sequences from extinct species such as Neanderthals, Denisovans, woolly mammoths, ancient zebras, penguins, giant sloths, and ancient sea cows. This work has generated a library of more than a million genetic recipes and yielded resurrected compounds with names like mammuthusin-2, mylodonin-2, and hydrodamin-1, illustrating how biological “code” can be algorithmically explored for antimicrobial, antimalarial, or anticancer potential.
Artificial intelligence is reshaping antibiotic discovery beyond de la Fuente’s lab, with groups led by James Collins at MIT and Jonathan Stokes at McMaster University moving from predictive models that screen known compounds to generative systems that design new molecules. De la Fuente’s team recently used one generative artificial intelligence model to design synthetic peptides and another to evaluate them, then tested two compounds in mice infected with a drug resistant strain of Acinetobacter baumannii, where both successfully and safely treated the infection. The field remains in a discovery phase, so de la Fuente is now building ApexOracle, a multimodal model that integrates chemistry, genomics, and language to analyze pathogens, identify genetic vulnerabilities, match them to antimicrobial peptides, and predict how peptide based antibiotics would perform in laboratory assays. Although still preliminary, these tools have already saved decades of research time and aim to convert vast biological diversity into practical treatments that can slow or outpace the evolution of resistance.
