Research On ATP-binding Proteins Prediction Based On Deep Learning

Proteins are important materials that compose all tissues and cells of the organism.Various biological processes in an organism require the participation of proteins.With the development of sequencing technology,many protein sequences have been sequenced.Identifying the functions of these proteins is an important task of life science research.Using traditional experimental methods to identify protein functions has high accuracy and reliability,but this often takes several months,which does not satisfy the requirements of protein recognition in the post-genomic era.Therefore,predicting protein functions based on computational methods has become the focus of life science research.This article discusses the topic of protein function predictions and conducts in-depth research on the prediction of the ATP-binding protein and its subclass ABC transporters based on deep learning.The specific research contents are as follows:1.We propose the ATP-binding protein prediction method Deep RCI,which obtains the evolutionary information of the protein by statistical calculation from the multiple sequence alignment of the protein to further obtain the spatial contact relationship between different residues in the protein sequence,and compare these residues The contact information is saved as a 400×400 image.This method uses a13-layer convolutional neural network as a classification model to predict ATP-binding proteins.The accuracies on the validation set and independent test set are 92.89% and 93.19%,respectively,compared with the existing ATP-binding prediction method improves the accuracy by 12.06%.In this thesis,Deep RCI was used to predict the proteins of unknown function in 2000 wild soybeans in Inner Mongolia,and 262 ATP-binding proteins were obtained.2.We propose Deep RTCP,a prediction method for ABC transporters.This method uses 7 kinds of pseudo-amino acids instead of the traditional 20 kinds of amino acids,which reduces the dimensionality and sparseness of amino acid composition features.The method combines the pseudo-amino acid-based tripeptide feature and the position-specific scoring matrix into an 80-dimensional composite feature and uses a 7-layer one-dimensional deep convolutional neural network as the classification algorithm.The accuracy of the 10-fold cross-validation of this method is 98.29%,which is 9.29% higher than the existing ABC transporter prediction method.