DeepMind announced, has successfully solved the protein folding structure that has been difficult for scientists to accurately predict for the past 50 years through its artificial intelligence technology.
Proteins are the fundamental building blocks of biological function, and different folding structures enable proteins to perform different biological functions. Therefore, research on proteins has been ongoing for a long time, hoping that by understanding protein folding structures, we can develop more antidotes to improve diseases or solve more protein-related problems.
However, the folding structure of proteins is quite complex, with at least 10 to the power of 300 possibilities, making it difficult to analyze using general calculation methods.
Precisely because it's difficult to analyze, many scientists are eager to solve this problem. The industry has also developed many research challenges, such as the previously promoted Folding@home project, which aims to study protein folding structures by allowing users to contribute their idle computing power and form a massive computing resource through networked collaboration. Even NVIDIA has previously invested in this research using its GPU computing resources.
DeepMind's successful deciphering of protein folding was primarily based on the AlphaFold design proposed in 2018. Through training with existing data on 17 protein structures and accelerated using 100 to 200 GPUs, the team was able to accurately predict protein folding after spending several weeks analyzing and learning. This achievement was recognized by the CASP experiment, a key test of protein structure prediction technology.
In the future, DeepMind plans to apply AlphaFold to various disease studies related to protein structure, and hopes to make contributions to the development of biology.
