OpenAI Profiles Lab Using ChatGPT and Codex to Speed Antimicrobial Search
César de la Fuente’s lab says its AI systems can shrink the first candidate search from years to hours. The harder work—showing a molecule is safe, effective and manufacturable—still happens in the lab.
Listen to this story
The audio brief
Story brief
3 key pointsOpenAI is highlighting César de la Fuente’s lab as an example of AI-assisted antimicrobial discovery, using the lab’s deep-learning systems alongside ChatGPT and Codex. The tools help researchers handle biological datasets, develop hypotheses, write code, and connect multidisciplinary findings, potentially compressing early candidate screening from years to hours. The commercial and medical payoff remains distant:...
- 01
The lab searches genome and protein data from living and extinct organisms for antimicrobial candidates.
- 02
ChatGPT and Codex support research workflows; they do not replace the lab’s specialized biological discovery models.
- 03
Candidates still require efficacy, dosage, human-cell safety, resistance, stability, manufacturing, and clinical validation.
The search for a possible antimicrobial molecule can take years, but César de la Fuente’s lab says its deep-learning systems can narrow that initial hunt to hours. In a newly published profile, OpenAI describes how the group combines those in-house models with ChatGPT and Codex to search biological data for candidates that might help fight drug-resistant infections.
The work is aimed at a large and urgent search space. The lab examines genome and protein datasets from living and extinct organisms, looking for molecules that could become antimicrobials. Its models are trained to recognize patterns in biological sequences, helping researchers sift through far more possibilities than a conventional sample-by-sample search.
Finding candidates is not making medicines
That distinction is central to the profile. A computationally promising molecule must still be tested to see whether it kills the target microbe, what dose is effective, and how it affects human cells. Researchers may also need to improve its effectiveness, safety or stability, assess whether microbes readily develop resistance, and establish a reliable manufacturing process before regulatory review and clinical trials.
“Ground-truth experiments are essential to validate AI predictions.”
César de la Fuente, in OpenAI’s profile
AI as a bridge across the lab
The profile presents ChatGPT and Codex less as replacements for the lab’s specialized discovery models than as general-purpose tools around the research workflow. Lab members use them to brainstorm hypotheses, write and refine code, process datasets, analyze results and connect ideas from biology, chemistry, computing and engineering.
Where the tools fit
- Downloading, organizing and preprocessing genome datasets.
- Helping researchers review unfamiliar biological, chemical and computational subjects.
- Supporting hypothesis development and code work across a mixed-discipline team.
De la Fuente describes ChatGPT as a collaborative sounding board: a place where researchers can develop a hypothesis while bringing together perspectives from people working on different parts of the same problem. He also cautions that AI output must be checked for accuracy, a constraint that matters when an early research suggestion could influence what a team chooses to test next.
A faster start against a growing threat
The appeal of speeding the front end of discovery is clear. OpenAI’s profile cites an estimate that bacterial antimicrobial resistance was associated with about five million deaths in 2021 and says the annual toll is projected to roughly double by 2050. But the profile does not present an AI-selected molecule as an approved treatment; it describes systems for prioritizing candidates for the experimental pipeline.
Sources
- openai.comHow a researcher uses Codex and ChatGPT to search for new antimicrobial molecules
Loading discussion...
Reader comments
Newest comments first. Replies stay oldest first.