
A team of researchers that included Biologists, AI experts, and Evolutionary experts at EvolutionaryScale and Arc Institute designed and built an AI model capable of generating code to synthesize new proteins through simulations. The paper that was published in the Science Journal, describes the factors that developed the ESM3 AI model, and how it was used to generate a bright protein.
Prior research suggests that the process of synthesizing proteins can provide previously unknown insights into the structure and function of natural proteins.
As of right now, most proteins are copies of natural proteins. Taking this principle into account, researchers used the ESM3 AI model to simulate the evolutionary process of a synthetic protein that does not naturally exist.
Generating artificial proteins opens a variety of previously unknown possibilities
Multimodal generative AI, like ESM3, can unlock new avenues of research, helping to better understand the nature of proteins and their uses. The fact that ESM3 is a multimodal language means that it is a chatbot, very much like the ones currently in everyday use, such as ChatGPT and Google Gemini.
The model also learns about the nature of things while being tried on massive banks of data. The ESM3 model was trained in 771 billion tokens, which were in turn generated from 3.15 billion protein sequences, 236 million protein structures, and 539 million protein annotations.
We're thrilled to present ESM3 in @ScienceMagazine. ESM3 is a generative language model that reasons over the three fundamental properties of proteins: sequence, structure, and function. Today we're making ESM3 available free to researchers worldwide via the public beta of an API… pic.twitter.com/BY4Glad59x
— Alex Rives (@alexrives) January 16, 2025
Researchers behind the simulation claim this was the equivalent of giving the AI model 500 million years of evolutionary knowledge. This allowed the model to start with basic code, which evolved in virtual time into a digital protein. It was then turned into a real-world synthetic protein using standard techniques. The physical result was a protein with a genetic sequence that was different from other existing proteins.
Additionally, the team specifically asked the model to generate a green fluorescent protein, which would shine under ultraviolet light.
AI simulations could have revolutionary implications for the study of proteins
The use of AI in protein design projects, much like the one carried out with the ESM3 model, is tipped to have profound real-world implications in the near future. For instance, AI-driven protein design could facilitate the creation of new proteins that would allow for the development of new medicines and vaccines.
AI tools have already allowed scientists to design proteins with specific functions, such as enzymes that can degrade environmental pollutants.
ESM3 generating a fluorescent protein! pic.twitter.com/eOPZmrYVm7
— Alex Rives (@alexrives) January 16, 2025
New synthetic proteins could also allow scientists to create new biotechnology materials, that could have a variety of applications. For instance, new biomaterials could be used for medical implants new sustainable materials, and the creation of complex nanostructures.
See all the latest news from Greece and the world at Greekreporter.com. Contact our newsroom to report an update or send your story, photos and videos. Follow GR on Google News and subscribe here to our daily email!


