OpenAI tools support faster antimicrobial candidate discovery

AI narrows the search for antimicrobial molecules

OpenAI says bioengineer César de la Fuente and his laboratory are using Codex, ChatGPT and in-house deep-learning models to search genomes and protein datasets for potential antimicrobial molecules. The work targets early discovery, where the lab says AI can reduce an initial search for candidate molecules from years to hours.

The research addresses drug-resistant microbes, including bacteria, fungi, parasites and viruses. About five million deaths in 2021 were associated with bacterial antimicrobial resistance, and the annual toll is projected to roughly double by 2050. De la Fuente noted that no new class of antibiotics has emerged for 50 years.

Biology treated as an information system

Rather than concentrating only on modifications to existing medicines or familiar chemical classes, the lab searches biological sequence data from living and extinct organisms. Its premise is that nucleotides, amino acids, proteins and peptides can be examined as information, enabling models to recognize sequence patterns associated with functional molecules.

Digital genome and protein databases make it possible to search broadly across the tree of life, but they also create a signal-selection problem. The lab’s models scan large datasets, identify patterns that may be difficult to detect manually, and prioritize a manageable group of molecules for experimental testing.

Codex and ChatGPT complement those models. Lab members use them to brainstorm hypotheses, write and refine code, download and prepare large genome datasets, analyze results, clarify unfamiliar terminology and compare methods across disciplines. This use of Codex builds on Codex development tools in ChatGPT by showing how development assistance can extend beyond conventional software tasks into computational biology workflows.

AI predictions remain an early research step

A promising sequence is not automatically a medicine. Researchers must establish whether a molecule kills the target microbe, determine an effective amount and assess its effects on human cells. Chemists may optimize effectiveness, safety or stability before further work examines toxicity, resistance, movement through the body and manufacturing.

Even candidates that pass these stages must undergo regulatory review and clinical trials before becoming approved antimicrobial drugs. De la Fuente stresses that ground-truth experiments are essential to validate AI predictions and that researchers must double-check AI output for accuracy.

For businesses applying AI to scientific or technical discovery, the practical implication is clear: use models to organize vast search spaces and support specialist workflows, while keeping domain experts and real-world validation responsible for consequential decisions.