Researchers detail how ProGen, an LLM trained on 280M proteins, designed proteins with anti-microbial properties that were tested in real life and shown to work
New ScientistKarmela Padavic-Callaghan
Context & Ripple Effects
This story closes a loop that opened in 2020, when MIT researchers used AI to find potent antibiotics against drug-resistant diseases — that was discovery by screening existing compounds. ProGen is a step further: a language model trained on 280M protein sequences that *generates* novel antimicrobial proteins, which were then physically synthesized and shown to work.
The arc matters because validation, not generation, has been the bottleneck for computational biology — ProGen is an early proof that LLM-style generative design can produce molecules that survive contact with the lab, a template the field later pushed all the way to AI-designed viruses that infect bacteria in 2026.
First-order effects
Antimicrobial research gains a generative design tool with experimental proof: researchers can now specify a desired property and synthesize candidates rather than only screen natural or existing compounds.
Second-order effects
Drug-discovery pipelines shift economics from screening libraries toward sequence-generation plus validation, pressuring biotech and pharma to build in-house protein language models or partner for them.
Third-order effects
If generative design scales from single proteins to full genetic systems — as the later AI-designed bacteriophages suggest — the same capability raises biosecurity questions alongside therapeutic ones, likely drawing regulatory attention to sequence-generation tools.
The trend: Biology is moving from AI-assisted discovery of existing molecules to AI-generated ones, with each validated design — antibiotics, proteins, viruses — expanding what generative models are trusted to specify.
“Here we describe ProGen, a language model that can generate protein sequences with a predictable function across large protein families, akin to generating grammatically and semantically correct natural language sentences on diverse topics.” https://www.nature.com/...
“An AI was tasked with creating proteins with anti-microbial properties. Researchers then created a subset of the proteins and found some did the job” — AI has designed bacteria-killing proteins from scratch - and they work @newscientist https://www.newscientist.com/ ...
ChatGPT for biology? Excited to share our work on LLMs for protein design out today @NatureBiotech https://nature.com/... + Proud to publicly announce @ProfluentBio with a $9M seed round to tackle meaningful challenges in biology with AI. Join us!
Using large language models like #ChatGPT for life science: making proteins from scratch via ProGen https://www.nature.com/... @NatureBiotech @salesforce @thisismadani @nikhil_ai and colleagues @SFResearch https://twitter.com/...
AI to create new members within protein families! Great! Also, important to mention that not all families are characterized, and within families, there is a lot of untapped subfunctionalization. Genomics enzymology approaches can help explore unknowns. https://www.nature.com/...