META AI has developed a competitor DeepMind AlphaFold

META AI has developed a competitor DeepMind AlphaFold

Researchers META AI have released the “Protein Language Model” ESM-2 with 15 billion parameters and the ESM METAGENOMIC ATLAS database containing more than 600 million prognostic structures of metagenomic compounds.

The system processes the sequences of genes using the method of self -controlled learning called “masked language modeling”.

According to scientists, they trained the algorithm on the array of sequences of millions of natural proteins.

“With this approach, the model should correctly add words in the passage of the text, for Market depth chart example,“ so that __ or not __, that is, __ ”. We have taught the language model to fill in passes in the sequence of proteins like “Gl_kke_ahy_g” among millions of different compounds, ”the study says.

ESM-2 is the largest and most effective neural network in its kind. According to scientists, the algorithm is 60 times faster than other modern systems like Deephafold from DeepMind.

The algorithm helped create ESM Metagenomic Atlas, predicting 617 million structures from the MGNIFY90 database in just two weeks of work on a cluster of 2000 graphic processors. To simulate the connection of 384 amino acids on one NVIDIA V100 video card, 14.2 seconds will be required.

“With modern computing tools, the prediction of the structure of hundreds of millions of proteins can take years, even using the resources of a large research institution. To make forecasts on the scale of metagenomics, the breakthrough in the rate of forecasting is crucial, ”the developers said.

Meta AI hopes that ESM-2 and ESM Metagenomic Atlas will advance science and help specialists studying the history of evolution or fighting diseases and climate change.

“We also study the methods of using language models to develop new proteins and promote the solution of problems related to health and the environment,” scientists added.

Recall that in July, the Alphafold algorithm DeepMind predicted almost all the compounds known in science found in plants, bacteria and animals.

In the same month, researchers from MIT developed a model of deep learning Equibind, which is 1200 times faster than analogues connects molecules with proteins when creating drugs.

In July 2021, artificial intelligence from Deepmind simulated 20,000 human protein structures.

Subscribe to FORKLOG news at Telegram: Forklog AI – all news from the world of AI!

Leave a Reply

Your email address will not be published. Required fields are marked *