Drug-resistant microbes—bacteria, fungi, parasites, and viruses—are a growing global threat. About five million deaths in 2021 were associated with bacterial antimicrobial resistance, a toll projected to roughly double by 2050. It can take years to find molecules with antimicrobial potential, and no new class of antibiotics has been introduced in 50 years. Much of modern development focuses on modifying existing medicines, an approach with diminishing returns.

César de la Fuente, a bioengineer, treats biology as an information system: nucleotides in DNA and amino acids in proteins and peptides are like an alphabet. His lab trains deep-learning models to recognize patterns in biological sequences, enabling searches across vast genome and protein datasets for potential antimicrobials. This can reduce the initial search from years to hours. Alongside their own models, the lab uses ChatGPT and Codex to brainstorm hypotheses, write and refine code, process datasets, analyze results, and connect ideas across scientific disciplines.

Only a fraction of a genome has a clearly understood function, and fewer sequences encode molecules that fight infectious microbes. Scientists have traditionally searched plants, animals, microbes, insects, water, and soil, collecting samples and testing candidates iteratively over years. Digital genome and protein databases now expand the search across the tree of life, but the abundance of data creates a needle-in-a-haystack problem. AI is well suited to scanning huge datasets, identifying patterns researchers might miss, and prioritizing a manageable set of candidates for experimental testing.

Finding a candidate is only the first step. Researchers must confirm it kills the target microbe, determine the effective amount, and test its effects on human cells. Chemists may optimize the molecule for effectiveness, safety, or stability. Further tests assess toxicity, how readily microbes develop resistance, and how the candidate moves through the body, along with reliable manufacturing methods. Candidates that pass these hurdles still face regulatory review and clinical trials. De la Fuente stresses that ground-truth experiments are essential to validate AI predictions, especially as the field continues to explore biology’s complexity.

The lab is highly transdisciplinary, with biologists, chemists, computer scientists, and engineers. Some members are strong programmers but know less about biology or chemistry, and vice versa. Codex and ChatGPT help bridge these gaps, allowing biologists to build programs and programmers to tackle biological problems. The tools also help lab members review unfamiliar topics, clarify terminology, compare methods across fields, and organize ideas. ChatGPT lets members work in their native languages, lowering barriers to accelerate workflows. De la Fuente uses ChatGPT as a brainstorming partner and describes the shared workspace as a collaborative sounding board fed by good and bad ideas from people who think differently. He cautions that AI outputs must always be double-checked for accuracy.

De la Fuente sees breakthroughs waiting at the edges between scientific disciplines, where few people go, and views AI as part of a longer tradition of tools that help researchers understand the world. Just as the telescope illuminated the cosmos and the microscope revealed the invisible, machines now help researchers understand, predict, and engineer biology.