DeepMind, Google’s artificial intelligence unit, has taken a big step towards predicting protein structure. DeepMind’s AlphaFold system has solved a key “protein folding problem” and reduced the computing time needed to solve the problem from months to hours, the company said, helping speed drug discovery and potentially solving a problem similar to mapping the human genome. AlphaFold can predict the shape of proteins at the atomic level. The breakthrough will help scientists design drugs and understand diseases.
Different levels of protein folding determine how it interacts with other molecules, and understanding structural changes in proteins is important for discovering how viruses such as COVID-19 invade human cells, designing enzymes to break down pollutants and improving crop yields. AlphaFold is the name of the protein research team. DeepMind explained that Fold is a scientific technique that uses AI to break down the folds of proteins.
DeepMind’s most famous AI product is AlphaGo, which scored 1 in 2016: 4 beat Lee Se-dol, the South Korean go champion for the first time, and later defeated the top Players from China, Japan and South Korea in the “online fast chess match”, with a score of 60-0-1. Even Ke Jie, the world’s no. 1 go player, was defeated.
As a result of the COVID-19 outbreak, DeepMind’s other AI team, AlphaFold, has applied AI techniques such as neural network models and deep reinforcement learning to the structural analysis of unknown proteins. This time, it has conducted extensive basic research on novel Coronavirus and opened up the available genome data. The genomic data could allow researchers to quickly develop test treatments for the virus.
DeepMind said AlphaFold predicts physical properties using neural networks that have been trained to predict the properties of a protein from its genetic sequence, such as the distance between pairs of amino acids or the angles between the chemical bonds that connect those amino acids, tweaking the structure to find the most efficient amino acid configuration. The first prediction used more time. It took the program two weeks to predict the first protein structure. Now it can predict it in a few hours.
