[ad_1]
Predicting protein structures from amino acid sequence information alone, known as the “protein folding problem”, has been an important open research question for over 50 years. In fall 2020, DeepMind’s neural network model AlphaFold took a big step forward in solving this problem, surpassing some 100 other teams in the Structure Prediction Critical Appraisal Challenge (CASP). , considered to be the benchmark precision assessment for predicting protein structure. . The success of the new approach is considered an important step in the prediction of protein structure.
This week the DeepMind Journal Highly accurate prediction of protein structure with AlphaFold has been published in the prestigious scientific journal Nature. The article presents AlphaFold2, a completely redesigned and open source model that can predict protein structures with atomic-level precision.
Although machine learning researchers have long sought to develop computational methods to predict 3D protein structures from protein sequences, there had been limited progress along this path, primarily due to the computational complexity of molecular simulation, context dependence on stability, the difficulty of producing sufficiently precise models for protein physics.
In this work, the DeepMind team presents the first computational approach capable of predicting the structures of proteins with near-experimental precision. The proposed AlphaFold2 model achieved “outstanding” results in the recent CASP14 assessment.
AlphaFold2’s realizations are based on neural network architectures that jointly integrate multiple sequence alignments (MSA) and pair functionality. The AlphaFold network can directly predict the 3-D coordinates of all heavy atoms for a given protein using the primary amino acid sequence and aligned sequences of homologs as inputs. The network consists of two main modules: Evoformer and a structure prediction module.
Evoformer views the prediction of protein structure as a graphical inference problem, representing the data as a graph in which the nodes represent pairs of amino acids and the edges as the proximity of these pairs to each other. compared to others in protein. By applying deep learning techniques, Evoformer gradually refines a prediction of what the protein backbone should look like, and then feeds the results of the prediction to the structure prediction module.
The structure prediction module performs a series of geometric transformations to further refine the shape of the protein for greater accuracy. The abstract 3D protein images in this module appear as twisted, ribbon-like loops that branch out from the main protein backbone.
As described at CASP14, the methodological advances of AlphaFold2 include: 1) the departure of multiple sequence alignments (MSA) rather than more processed features such as inverse covariance matrices derived from MSAs, 2) replacement of 2D convolution by an attention mechanism that better represents interactions between distant residues along the sequence, 3) Use of a two-track network architecture in which the information at the level of the 1D sequence and at the level of the 2D distance map are iteratively transformed and transmitted back and forth, 4) Using an SE (3) -equivariant transformer network to directly refine atomic coordinates (rather than 2D distance maps as in previous approaches ) generated from the two-track network, and 5) End-to-end learning in which all network parameters are optimized by backpropagation from from the final result generated 3D coordinates through all the layers of the network to the input sequence.
AlphaFold has now clearly demonstrated its effectiveness in this important and rapidly evolving area of research, and DeepMind believes the model and associated computational approaches that apply its techniques to other biophysical problems may soon become essential tools in cutting-edge research. in biology.
The AlphaFold2 code is available on the Github project. The paper Highly accurate prediction of protein structure with AlphaFold is about Nature.
Author: Hecate Il | Editor: Michael Sarazen, Zhang Channel
We know you don’t want to miss any news or research breakthroughs. Subscribe to our popular newsletter Weekly Synchronized Global AI to get weekly AI updates.
[ad_2]
Source link