César de la Fuente’s lab is using OpenAI’s Codex and ChatGPT to search living and extinct genomes for molecules that could fight drug-resistant infections, according to OpenAI’s labs team. The work targets one of medicine’s most urgent problems: bacteria that no longer respond to existing drugs. What stands out here is the method. Instead of testing compounds one by one in a lab, the team mines biological data at scale, using AI to surface candidates worth synthesizing.
This matters because antimicrobial resistance kills more than a million people a year, and the pipeline for new antibiotics has been thin for decades. De la Fuente’s approach reframes the problem as a search problem. If the answers are already hidden in genetic sequences, including those of extinct organisms, the job is to find them fast.
Here’s how a workflow like this comes together, and how you can borrow the thinking for your own data-heavy research.
Quick Start
You’ll learn how a research lab pairs Codex and ChatGPT to scan massive biological datasets for useful molecules. You don’t need a biology degree to take the lesson away. You need a large dataset, a clear target, and a willingness to let AI write and run the code that filters it.
- Define what you’re looking for. Start with a precise target. In this case, it’s antimicrobial peptides, short protein sequences that can kill bacteria. A sharp definition matters because it shapes every filter downstream. Vague goals produce noise. Specific goals produce candidates you can actually test.
- Gather the raw data. The lab pulls from genomes of living and extinct organisms. That second part is the clever move. Extinct and ancient genetic material is an untapped library, free from the bias of what modern science has already studied. The lesson: look where others aren’t looking.
- Use Codex to build the search tools. Codex writes and runs the code that parses genomes and flags sequences matching the target profile. This is where AI replaces weeks of manual scripting. You describe the filter you want in plain language, Codex produces the code, and you iterate. The point is speed. You test ideas in hours, not sprints.
- Use ChatGPT to reason through results. Once candidates surface, ChatGPT helps interpret them, explain patterns, and decide what’s worth pursuing. Think of it as a research partner that never tires of your follow-up questions. It won’t replace lab validation, but it sharpens which molecules earn a spot on the bench.
- Validate in the lab. AI narrows the field. Real experiments confirm whether a candidate actually works. This step is non-negotiable. Every AI-flagged molecule is a hypothesis, not a result, until it’s tested against real bacteria.
Tips worth keeping
- Treat AI as a filter, not an oracle. It ranks possibilities; you make the call.
- Feed it clean, well-structured data. Garbage in still means garbage out, even with strong models.
- Keep a human in the loop at every decision point. The stakes in drug discovery leave no room for blind trust.
Why this signals a bigger shift
De la Fuente’s lab is a preview of how AI is changing scientific research broadly. The bottleneck used to be human time spent writing code and sifting data. Tools like Codex collapse that bottleneck, letting scientists spend their hours on judgment and experiments instead. The same pattern applies far beyond biology. Any field sitting on huge, underused datasets can run a version of this playbook.
Your next steps
If you work with large datasets, identify one search problem you’ve been avoiding because the coding felt too heavy. Describe it to Codex and let it draft the tooling. Use ChatGPT to interpret what comes back. Then, critically, verify the output against ground truth before you act on it. Start small, prove the loop works, and scale from there.
Full details on the lab’s work are available at the original source.