The First New Antibiotic Class in 50 Years, With Codex and ChatGPT on the Case!

Hi, it's me, Shiichan! Today's a story where science and AI team up in a genuinely moving way.

OpenAI News openai.com

What was announced?

OpenAI News published a story about how the César de la Fuente lab uses ChatGPT and Codex to search for new antimicrobial candidates. What makes it unique is that they search not just living organisms, but genomes of extinct ones too.

Why it matters

Antimicrobial resistance is a global threat — an estimated 5 million deaths in 2021 alone were associated with it. And yet no new class of antibiotics has been developed in 50 years. Traditional approaches are losing effectiveness, making new discovery methods urgent.

What changes

The de la Fuente lab treats biology as an information system: DNA bases and protein amino acids are, in their framing, "like an alphabet." Deep learning models recognize patterns in biological sequences and search genome and protein databases for novel antimicrobial candidates. This approach has cut the initial search for candidate molecules from a matter of years down to hours.

Dive Deep

The lab uses ChatGPT and Codex for:

• Generating and testing hypotheses
• Writing and refining code
• Processing datasets
• Analyzing results
• Bridging knowledge across disciplines

Many lab members come from different academic backgrounds, and AI plays a role in lowering the barrier between scientific disciplines — letting programmers engage directly with biological problems.

The researchers are careful to note that accuracy always needs to be double-checked, and they're clear-eyed about the risk of relying on AI alone. In their view, an AI prediction only becomes reliable once it's been validated experimentally in the lab. Their stance is that progress in AI and progress in lab biology need to move forward together — neither one works alone.

Wrap-up

• OpenAI News shared how the César de la Fuente lab uses ChatGPT and Codex to search for new antimicrobial candidates.
• Antimicrobial resistance is linked to millions of deaths a year, yet no new antibiotic class has emerged in 50 years.
• Treating DNA and proteins as information for deep learning cut candidate search time from years to hours.
• ChatGPT and Codex are used broadly, from hypothesis generation through coding to result analysis.
• Their insistence that every AI prediction be experimentally verified is a striking detail — well worth reading if you're interested in AI-science collaboration.