Prediction method and device of protein signal peptide based on geometric graph neural network

Abstract

The present invention relates to a method and device for predicting protein signal peptides based on a geometric graph neural network. The method includes: acquiring protein sequences in a data set, extracting character sequences of signal peptides; performing feature extraction on the signal peptides to obtain the The sequence evolution features, physical and chemical features, structural features, and statistical features of the signal peptide; the evolutionary features, physical and chemical features, structural features, and statistical features are fused to obtain a multidimensional vector; and the signal is constructed according to the character sequence and the multidimensional vector. The feature map of the peptide; using the feature map as the input of the geometric graph neural network model, training the geometric graph neural network until its error is lower than a threshold, saving the geometric graph neural network model and using it to predict the signal to be predicted peptide. The invention combines the geometric graph neural network and the character sequence, reduces the over-fitting of the model while retaining the original information of the feature extraction, and improves the accuracy of signal peptide prediction.

Description

technical field [0001] The invention relates to the fields of biological information and deep learning, in particular to a method and device for predicting protein signal peptides based on geometric graph neural networks. Background technique [0002] In organisms, most proteins do not function in the form of monomers, but perform different biological functions in the form of interactions. Among them, protein-protein interaction (Protein-Protein Interaction, PPI) refers to the process of forming a protein complex by two or more molecular proteins through covalent bonds. Protein interactions play important roles in most biochemical functions. For example, signaling molecules interact with proteins to transmit signals from outside the cell to the inside of the cell, and signal transmission is the basis for many functions. For protein interaction, in essence, it is realized through the mutual binding of some residues on the protein, and these residues are called protein-prote...

AI Technical Summary

Problems solved by technology

[0005] The present invention aims at the problems of overfitting and inability to accurately identify the cleavage site existing in the existing methods for predicting protein signal peptides, and provides a prediction of protein signal peptides based on a geometric graph neural network in the first aspect of the present invention The method comprises the following steps: obtaining the protein sequence in the data set, extracting the character sequence of the signal peptide; performing feature extraction on the signal peptide to obtain the sequence evolution characteristics, physicochemical characteristics, structural characteristics, and statistical characteristics of the signal peptide; Fusing evolutionary features, physicochemical features, structural features, and statistical features to obtain a multidimensional vector; constructing a feature map of a signal peptide according to the character sequence and the multidimensional vector; using the feature map as an input to a geometric graph neural network model , train the geometric graph neural network until its error is lower than a threshold, save the geometric graph neural network model and use it to predict whether the amino acid sequence to be predicted contains a signal peptide fragment

Images