How one scientist leverages Codex and ChatGPT to discover novel antimicrobial compounds

How AI Tools Like Codex and ChatGPT Hunt for New Antibiotics

Meta description: Learn how César de la Fuente's lab uses Codex and ChatGPT to mine living and extinct genomes for peptides that fight drug-resistant bacteria.

Tags: AI in science, antibiotics, drug discovery, Codex, ChatGPT

Featured image: A friendly abstract illustration of a glowing DNA helix dissolving into small floating molecules, with a soft palette of teal and warm yellow.

For researchers fighting drug-resistant bacteria, help may come from an unexpected source: language models. A team led by César de la Fuente is applying Codex and ChatGPT, tools best known for writing code and conversation, to a very different problem: scanning genomes, including those of long-extinct organisms, in search of molecules that could become the next generation of antibiotics. This article explains how the approach works, why ancient DNA matters, and what it could mean for modern medicine.

Why new antibiotics are so hard to find

Antibiotic resistance is one of the biggest threats in global health. Bacteria evolve quickly, and many common infections are becoming harder to treat with existing drugs. Traditional antibiotic discovery is slow and expensive: scientists screen huge numbers of compounds, and only a tiny fraction ever show promise. That is where artificial intelligence changes the picture. Instead of testing molecules one by one in the lab, researchers can use AI models to sift through enormous amounts of biological data and flag the most promising candidates first.

Treating biology like a language

The core insight behind the lab's work is that DNA and proteins can be read like text. Just as a language model predicts the next word in a sentence, it can learn patterns in sequences of amino acids, the building blocks of proteins. De la Fuente's team uses Codex and ChatGPT to look for short protein fragments, known as antimicrobial peptides, that show signs of being able to kill bacteria.

Rather than guessing randomly, the models learn from known examples what an effective antimicrobial sequence tends to look like. They can then comb through genome after genome, spotting candidate peptides that researchers might never have thought to test. The lab applies this method to the genomes of living organisms and extinct species alike, dramatically widening the pool of possible drug leads.

Why extinct genomes matter

One of the most intriguing parts of this research is the use of ancient DNA from extinct organisms. Molecules preserved in species that disappeared thousands or even millions of years ago may have chemical properties that modern organisms no longer carry. Because today's bacteria have never encountered these ancient peptides, they may have no built-up resistance to them. In other words, molecules from the distant past could offer a fresh angle against bacteria that have learned to shrug off current drugs.

What this means for the future

The work does not mean an AI can simply invent a finished antibiotic overnight. Candidates flagged by the models still need to be synthesized and tested in the lab to confirm they are safe and effective. But the approach can dramatically narrow down the search, saving time and resources at the earliest and most expensive stage of drug discovery. If methods like these continue to mature, they could become a standard part of how scientists respond to the growing crisis of drug-resistant infections, and a reminder that tools designed to handle language can also help us read the language written in our DNA.