Protein folding identification method based on triple loss

Abstract

The invention discloses a protein folding identification method based on triple loss, which comprises the following steps of encoding protein by using one-hot encoding, inputting the encoded protein into an SSA program to obtain a contact graph between protein residues, and using the contact graph as input data. inputting the input data into a pre-trained deep learning framework, wherein the output of the network is the characteristic that the protein is specific to folding identification, comparing characteristics of the query protein with template proteins of known protein folding categoriesin a protein database, and assigning the folding category of the template protein closest to the query protein to the query protein. According to the method, the training thought of triple loss is used for reference, so that protein structures of the same class are closer, protein structures of different classes are farther, feature expression of protein has higher discriminability, and the recognition efficiency is higher.

Description

technical field [0001] The invention belongs to the field of bioinformatics prediction of protein structure, in particular to a protein folding recognition method based on triplet loss. Background technique [0002] The determination of protein folding type can reveal the second set of genetic code of life, specifically how the primary structure of protein determines its spatial structure. As we all know, the three-dimensional structure of proteins plays a crucial role in the study of protein functions and properties, and the correct identification of protein folding is a key part of predicting the three-dimensional structure of proteins. In addition, because the protein folding method has a profound impact on protein heterogeneity and molecular function, it will have a huge role in promoting the artificial design of proteins in medicine, the search for fatal mechanisms, and the refolding of inclusion bodies. Therefore, fast and accurate identification of protein folding ty...

AI Technical Summary

Problems solved by technology

[0007] The deep neural network method can automatically extract features from the input data, which greatly improves the accuracy of fold recognition, but the current related methods can only use a series of samples of known fold types to train the deep neural network, and then use the middle layer Feature degeneralization for proteins of unknown fold type

The main disadvantage of this method lies in its indirectness and inefficiency: you can only pray that the features of the middle layer can generalize new proteins well, and the feature dimension of the middle layer is very large, which will cause "dimension disaster".

Images