Finding a molecule that might become an antibiotic is a needle-in-a-haystack problem, and the haystack keeps growing. In an OpenAI account of the work, bioengineer César de la Fuente describes a lab that treats biology as an information system: DNA nucleotides and protein amino acids work like an alphabet, and deep-learning models trained on biological sequences can scan genome and protein datasets for patterns that suggest antimicrobial activity. The supplied material says this can cut the initial candidate search from years to hours.
The bottleneck is not compute, it is cross-disciplinary translation
De la Fuente’s lab is deliberately transdisciplinary, with biologists, chemists, computer scientists, and engineers. Some members write strong code but know less biology; others are the reverse. Codex and ChatGPT sit in that gap. According to the source, lab members use them to review unfamiliar topics, clarify terminology, compare methods across fields, and organize drug-discovery ideas. Biologists can build programs; programmers can take on biological problems. That is a different value proposition from “AI finds drugs.” It is AI as a translation layer between specialists who would otherwise need years to overlap.
Where the tools sit in the pipeline
Alongside the lab’s own deep-learning models, ChatGPT and Codex are used to brainstorm hypotheses, write and refine code, process datasets, analyze results, and connect ideas across disciplines. Some lab members use AI to download, organize, and pre-process large genome datasets. ChatGPT also lets members work in their native languages, which the source frames as lowering barriers to faster workflows.
De la Fuente uses ChatGPT as a brainstorming partner and describes the shared workspace as a collaborative sounding board fed by people who think differently about the same problems — including bad ideas. He also cautions that accuracy always needs double-checking. That caveat matters for anyone building on model output: the lab’s own deep-learning models handle the sequence search, while general-purpose tools handle the surrounding reasoning and code.
Prediction is the cheap part
The source is explicit that identifying a promising candidate is not the same as producing a medicine. Candidates must be confirmed to kill the target microbe, dosed effectively, and tested against human cells. Chemists may optimize for potency, safety, or stability. Further tests cover toxicity thresholds, how readily microbes develop resistance, how the molecule moves through the body, and whether it can be manufactured reliably. Approved drugs still face regulatory review and clinical trials.
That is why de la Fuente argues AI and laboratory biology have to advance together. Ground-truth experiments validate predictions, and he calls that critical for the life sciences in the years ahead. For builders, the practical read is that AI compresses the search and the coordination overhead, not the validation. If you are designing agent workflows for research, the handoff to wet-lab confirmation is the part you cannot skip — and the part that determines whether the speedup is real.
What to take from this if you build research tools
Two things stand out. First, the highest-leverage use here is not autonomous discovery but lowering the cost of crossing disciplines, which is closer to the capability-building pattern in how FDEs build a team’s capability instead of its dependency. Second, the source does not specify which Codex or ChatGPT versions the lab uses, how many candidates have been validated, or what the hit rate looks like. Those details are absent from the supplied material, so treat the hours-versus-years framing as the lab’s characterization of its search stage rather than an end-to-end result. The honest next step for a builder is to instrument exactly where the model output gets checked against an experiment — that is where the value either holds or evaporates.
Sources
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.