AI-based platform for predicting and decoding protein structures
2021
DeepMind, London, United Kingdom
In order to optimize and accelerate the decoding and prediction of protein structures, the company DeepMind has developed the AI-based computer network AlphaFold. To predict protein structure, AlphaFold uses sequential as well as structural information from various data sets. The platform is based on a deep convolutional neural network trained with millions of known protein structures from the experimental Protein Data Bank (PDB). To develop a deep learning algorithm, the system analyses the primary amino acid sequence of a protein while taking into account evolutionary, physical, and geometric information that influences the spatial arrangement of atoms in a protein. Thanks to the integrated machine learning system, the system is able to independently identify missing contexts and optimizes the protein structure predictions in a continuous training process. The results of the present study show that by combining bioinformatics and physical approaches, AlphaFold can predict the three-dimensional structure of individual proteins extremely precisely and with almost experimental accuracy, and also enables a detailed representation of complex protein structures. The proof of concept of the method was carried out as part of the 14th Critical Assessment of Protein Structure Prediction (CASP14).
Highly accurate protein structure prediction with AlphaFold
John Jumper
Added on: 10-19-2023
[1] https://www.nature.com/articles/s41586-021-03819-2
