Google Deepmind published the source texts of the machine learning system Alphafold 3, designed for spatial modeling and predicting a three-dimensional structure of proteins. For the creation of machine learning algorithms implemented in the second version of Alphafold, this year the Nobel Prize in Chemistry was awarded. Alphafold 3 tools are associated with Python and C ++, and spread under the license CC by-nc-sa 4.0. Plasticized models are provided based on User Agreement . A server has also been launched to allow experimenting with Alphafold 3 in online mode.
The model, which is trained at collections with a description of the structures of all known proteins and amino acid sequences, solves the problem protein folding and allows you to predict the three-dimensional structure of new proteins with an accuracy that is not inferior to laboratory analysis. The third version of the AlphaFold model is distinguished by the use of the new architecture “Pairformer”, which develops the idea of the “