The AI startup EvolutionaryScale has released ESM3, a generative LLM with 98 billion parameters for “programming biology”.
The company's focus is on proteomics, the study of the interactions, functions, compositions and structures of proteins and their cellular activities.
While multimodal models like GPT-4 can generate text or images, ESM3 is an AI tool for prototyping and creating recent proteins.
When a ribosome makes a protein, it uses mRNA, which incorporates the code to make a selected protein.
Every living organism has the identical genetic code with the identical 20 amino acids. If you can read and understand that code, you can program the ribosome to make a protein when needed.
According to EvolutionaryScale, ESM3 understands all this biological data, translates it, and speaks it fluently for use as a generative tool.
Instead of a laborious and expensive trial-and-error process within the laboratory, ESM3 can predict the form and performance of a protein in a simulation.
We have trained ESM3 and are pleased to introduce EvolutionaryScale.
ESM3 is a generative language model for programming biology. In experiments, we found that ESM3 can simulate 500 million years of evolution to generate recent fluorescent proteins.
— Alex Rives (@alexrives) 25 June 2024
ESM3 is trained on billions of proteins present in nature. One of the most important challenges in constructing the model was tokenizing the three-dimensional protein structure and its functions.
To do that, a technique needed to be developed that might allow any three-dimensional structure and performance to be written as a sequence of letters using discrete alphabets.
After being trained on billions of proteins, ESM3 speaks the language of nature fluently and might reason in regards to the sequence, structure and performance of proteins.
To exhibit the capabilities of ESM3, EvolutionaryScale used it to create a novel green fluorescent protein (GFP). GFPs are answerable for the gorgeous fluorescence we see in some life forms like jellyfish or corals.
GFPs are incredibly rare in nature. The company estimates that the brand new protein, which it calls esmGFP, “corresponds to over 500 million years of natural evolution performed by an evolutionary simulator.”
EvolutionaryScale is making the ESM3 model publicly available and hopes it’ll “enable scientists to explore the frontiers of protein design and artificial biology and find recent solutions to a few of our world's most vital problems.”
The dual-use and open source nature of a tool like ESM3 poses potential risks that the corporate says it goals to mitigate with its Responsible Development Framework.
Using AI to predictably program biology could lead on to the event of proteins that sequester carbon, devour stubborn pollutants equivalent to plastics, or recent medicines.
Advances in AI with tools equivalent to ESM3, AlphaFold and CRISPR could soon result in the eradication of diseases and environmental problems which have challenged science for a long time.
