ChatGPT and Codex Are Helping One Lab Hunt for the Next Class of Antibiotics
Bacterial antimicrobial resistance killed roughly five million people in 2021, and that toll is projected to roughly double by 2050. Against that backdrop, one research lab is using ChatGPT and Codex alongside deep-learning models to search genomes for new antimicrobial candidates, compressing a search process that once took years into hours.
Bacterial antimicrobial resistance killed roughly five million people in 2021, and that toll is projected to roughly double by 2050. Against that backdrop, one research lab is using ChatGPT and Codex alongside deep-learning models to search genomes for new antimicrobial candidates, compressing a search process that once took years into hours.
Fifty years without a new class of antibiotics
César de la Fuente, whose lab is behind this work, points to a stark fact: no new class of antibiotics has been developed in fifty years. Meanwhile the pathogens keep adapting, and the death toll keeps climbing. The five million deaths associated with bacterial antimicrobial resistance in 2021 are not a static figure — projections put the annual toll at roughly double that by 2050. That trajectory is what makes the search for new candidate molecules urgent, and it's also what makes the speed of AI-assisted discovery worth paying attention to.
Turning genomes into a searchable space
The lab's approach starts with deep-learning models trained to analyze biological sequences for antimicrobial potential. Instead of limiting the search to organisms alive today, the lab scans genomes of both living and extinct organisms, treating ancient genetic material as a source of candidate molecules that evolution has already tested in some form. Sequences that show promise are flagged and ranked, giving researchers a shortlist to prioritize for experimental testing rather than an undifferentiated pile of raw data.
Where ChatGPT and Codex fit into the workflow
The models do the sequence analysis, but ChatGPT and Codex handle much of the surrounding work. Codex is used to write and refine the code that processes datasets and runs analyses. ChatGPT serves as a brainstorming partner for developing hypotheses, and lab members also use it to review unfamiliar topics and clarify terminology when a question falls outside their immediate expertise. It's also used to organize and preprocess large genome datasets before they ever reach the deep-learning models — a step that determines how usable the downstream analysis will be.
Why the shortcut still ends in the lab
None of this replaces the wet lab. The role of AI here is to narrow years of initial candidate search down to a matter of hours, producing a prioritized list of molecules worth testing — but those predictions still have to be validated experimentally before they mean anything. Researchers are advised to double-check AI-generated outputs for accuracy rather than take them at face value, whether that output is a piece of analysis code, a summarized dataset, or a hypothesis ChatGPT helped sketch out. The AI compresses the search; the lab still has to confirm what it finds.
Specs
- AI systems: ChatGPT and Codex
- Primary use: Antimicrobial candidate discovery
- Search input: Genome and protein datasets
Sources
- OpenAI — OpenAI