De la Fuente’s lab uses ChatGPT and Codex to hunt for antibiotics
In 2021, 4.71 million deaths involved drug-resistant bacteria. The Lancet’s GRAM study forecasts 8.22 million by 2050.
It has been about 50 years since a new class of antibiotic reached patients. Everything modern medicine treats as routine, surgery, childbirth, chemotherapy, rests on an antibiotic that still works.
César de la Fuente runs a lab at the University of Pennsylvania. He treats biology as an information system: the nucleotides that build DNA and the amino acids that build proteins are, to him, an alphabet.
The lab’s job is to search that alphabet for sequences that kill bacteria. The candidates it finds are synthesized and tested in human cells and animal models.
Codex takes the software work: writing scripts, parsing data, running batch jobs, trying exploratory code. These were the jobs that used to jam the pipeline.
ChatGPT works as the lab’s communal brain. De la Fuente says their workspace takes input from all these people who think differently: biology, chemistry, computer science and engineering end up talking in one place.
The search is not limited to living organisms. The lab has found new molecules in snake venom, woolly mammoths and extinct penguins. De la Fuente calls this molecular de-extinction.
More than 37,000 sequences have been found encrypted in extinct organisms. The logic is simple: nature’s dataset is finite, and dead genomes make it bigger.
A distinction matters here. The models that found the antibiotic candidates are the lab’s own. AMP-Diffusion generated 50,000 candidates, 46 were synthesized, and two matched levofloxacin and polymyxin B in mice with no adverse effects observed.
ChatGPT and Codex did not find those molecules. What they did was build the workflow that cut the candidate search from years to hours.
A candidate is not a drug. Turning a molecule found on a computer into an approved medicine is a long and expensive road, and most candidates fall off it. Toxicity, resistance, dosing, effects on human cells, each is a separate stage.
De la Fuente stresses the same thing: real experiments are essential to validate what the AI predicts.
Sources
- OpenAI, “How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules”, (openai.com)
- OpenAI, “Accelerating antibiotic discovery with ChatGPT”, (openai.com)
- The Lancet, “Global burden of bacterial antimicrobial resistance 1990–2021: a systematic analysis with forecasts to 2050”, (thelancet.com)
- Penn Engineering, “Penn Engineers Unveil Generative AI Model that Designs New Antibiotics”, (engineering.upenn.edu)
- Nature Biomedical Engineering (PMC), “Deep-learning-enabled antibiotic discovery through molecular de-extinction”, (pmc.ncbi.nlm.nih.gov)
- MIT Technology Review, “The scientist using AI to hunt for antibiotics just about everywhere”, (technologyreview.com)
About this story
This story was posted on Instagram by @jarrus.tech on Sept. 22, 2026.
Spotted an error in this story? [email protected] · Instagram
Short link: thejarrus.com/en/c7
This story in Turkish: De la Fuente’nin laboratuvarı antibiyotik ararken ChatGPT ve Codex kullanıyor
On the same topic
DeepMind-born drug discovery company Isomorphic Labs in funding talks at a $40-50 billion valuation
Isomorphic Labs, the DeepMind-born AI drug discovery company, is in early talks to raise money at a valuation of at least $40 billion, Bloomberg reported.
OpenAI is rolling out GPT-6 to everyone in ChatGPT
OpenAI began rolling out GPT-6 to everyone in ChatGPT on October 7. Its new Intelligent UI builds answers from text, visuals and interactive elements.
OpenAI’s unreleased model wrote 722 math papers
On October 6, OpenAI published 722 math papers from an unreleased internal model on GitHub. It withdrew three on October 7, leaving 719 in the catalog.