Deep Mind to publish database of nearly all proteins in 3D structure

Deep Mind to publish database of nearly all proteins in 3D structure

28. Juli 2021 0 Von Horst Buchwald

Deep Mind to publish database of nearly all proteins in 3D structure

San Francisco, 7/28/2021

DeepMind, Alphabet’s AI subsidiary, will publish the most accurate and complete database of the 3D structures of nearly all proteins in the human body, which could advance drug discovery, new medicines and more. The AlphaFold Protein Structure Database, developed using DeepMind’s AlphaFold 2 AI system, is designed to improve our understanding of how proteins work.

Last year, DeepMind said it developed „AlphaFold“ AI software that can predict in just a few days what structure proteins will „fold“ into. How proteins fold determines almost every biochemical reaction in the body, and knowing their shape helps researchers determine what they do and how effective certain drugs might be, for example.

Deepmind’s AIphafold was trained on ~170,000 proteins and their structures. It scored an average of 92.4 out of 100 on protein structure prediction.

Some experts have compared the potential impact to the Human Genome Project. „We believe this work represents the most significant contribution AI has made to the advancement of scientific knowledge to date,“ said DeepMind founder and CEO Demis Hassabis.

A 16-page abstract is available in the journal Nature, along with a 62-page collection of supplementary material and the code library on GitHub.

https://www.nature.com/articles/s41586-021-03819-2