Last year DeepMind presented AlphaFold v2, which predicts 3D structures of proteins down to atomic accuracy. Today they share the methods in their latest paper at Nature along with open source codes. It is inspiring to see the research this enables.
This new model, AlphaFold v2.0 has been published in Nature and entered into the CASP14 competition. AlphaFold is a complete pipeline for protein structure prediction that includes many improvements from previous iterations of its predecessor models
Deepmind has pushed the boundaries of computing. AlphaFold is a computational method that predicts protein structures with atomic accuracy even where no similar structure exists. They demonstrated this in the challenging 14th Critical Assessment of Protein Structure Prediction (CASP14).
AlphaFold is being developed by a team that includes both biologists and computer scientists. This new approach to designing the deep learning algorithm taps into our understanding of how proteins are structured to make it more accurate than other (less-informed) machine algorithms currently on the market today.
Here is DeepMind’s paper detailing how its AlphaFold 2 software is able to outperform other similar programs.
The code has been open-sourced for researchers worldwide and promises to be an invaluable tool in the future of medical research.