Scientists from EvolutionaryScale, a company created by former META researchers, introduced a new artificial intelligent model ESM3, which is capable of Create proteins from scratch. The model works on the principle of predicting sequences, similarly to how ChatGPT generates the text. The study was published on July 2 in the database biorxiv .
New opportunities for synthetic biology
Model esm3 allows you to develop proteins that are not found in nature, which opens up broad prospects for synthetic biology. In particular, scientists were able to create a new fluorescent protein, which is 58% different from natural analogues. This protein shines in a new shade of green and was called “Esmgpf”.
ESM3 is a large language model similar to the GPT-4 from Openai, and was trained in 2.78 billion proteins. The model removes information about the sequence, structure and function of proteins, and then predicts the missing fragments. Thus, ESM3 not only predicts existing proteins, but also generates new ones with specified functions.
In 2022, the team from META introduced the predecessor of ESM3- EMSFOLD , which predicted structures microbial proteins. In the same year, DeepMind announced its AlphaFold3 model, capable of predicting structures of 200 million proteins. However, all these models had restrictions and their predictions needed confirmation. H3> ESM3 advantages
ESM3 uses information About 771 billion unique proteins and is able to generate proteins with special functions. This allows you to significantly accelerate the process of searching and creating protein structures, which otherwise would be slow and expensive.
During the study, scientists requested a model to create a new fluorescent protein. ESM3 generated 96 proteins options, from which the most different from natural ones was chosen. This protein turned out to be 50 times less than natural analogues, but the subsequent iterations of the model led to the creation of a brighter protein “ESMGPF”.
ESM3 technology can find use in various areas, from the development of new drugs to the creation of chemicals to destroy plastic. The small version of the model is already available on a non -profit license, and the large will be provided to commercial researchers.
Conclusion
ESM3 development is a significant step forward in the field of synthetic biology and demonstrates the possibilities of artificial intelligence in the creation of new biological structures. This breakthrough can lead to important discoveries and innovations in various scientific and industrial fields.