Google's DeepMind division has released a number of unique and remarkable AI models, including one that can surpass you in StarCraft II (and almost anyone else). DeepMind, on the other hand, isn't just interested in AI for gaming. AlphaFold, a machine learning algorithm that can predict the shape of proteins, was unveiled by the company last year. DeepMind has now revealed that it has produced structures for all 200+ million proteins in the UniProt database. This is significant for basic biology research and efforts to address some of our day's most pressing scientific puzzles.
Proteins are the foundation of all biological life on Earth, but knowing the amino acid sequence of a protein does not mean you understand what it does or how it operates. A protein's sequence determines its positive and negative charge patterns, hydrophilic and hydrophobic regions, and cross-linked segments. This establishes the protein's active shape, or "conformation," as it is known in the lab, and the conformation of a protein is what gives it its activity. Even minor errors in structure prediction can be the difference between an enzyme that successfully catalyzes a process and one that accomplishes nothing at all.
Determining the conformation can be a time-consuming process that frequently relies on specialized techniques such as X-ray crystallography. With highly accurate conformation predictions, AlphaFold helps put that data into context. In the video below, a team from the University of Colorado, Boulder discusses the difficulties of investigating proteins implicated in bacterial antibiotic resistance. The team spent ten years debating the form of a protein that AlphaFold predicted in a matter of minutes. That's because AlphaFold has been trained on over 170,000 known protein structures, allowing it to anticipate how newer sequences will look in 3D.
DeepMind planned to make the AlphaFold database publicly available when it released AlphaFold last year. There were just one million structures available at the time, so the 200-fold rise in the last year is rather impressive. According to DeepMind, AlphaFold has been mentioned in over 4,000 scientific articles since its release, and it could help scientists grasp urgent challenges such as antibiotic resistance, food security, and the effects of plastic pollution.
DeepMind will present a predicted sequence immediately on the web page now that the entire UniProt database has been completed. A bulk download of the entire database of all 200 million structures will also be available via a Google Cloud Public Database.
0 Comments