Application Of Deep Learning Algorithm In Protein Structure Prediction

With the change of time,the technology of human genomics has been greatly developed.At the same time,the amount of protein sequence data and biological genetic data has also greatly increased.More and more scientific researchers have adopted different data Analytical methods are used to deal with these huge data.In this era of artificial intelligence and big data,mining various types of algorithms on a large amount of biological data has become a popular research method.In recent years,machine learning and deep learning algorithm technologies have been widely developed and applied in the field of bioinformatics,and have achieved some remarkable results,such as the problems related to protein structure prediction to be carried out in this research institute.In this research,a series of work was performed on predicting the structure of proteins based on deep learning algorithms.Using the protein structure predicted by deep learning algorithms can effectively provide technical support for biological experts to further study the function of proteins.The focus of research is to perfectly predict the multilevel structure of proteins,and then to be able to facilitate the exploration of protein functions through the predicted protein structure.Therefore,this study focuses on two important subproblems: the prediction of protein secondary structure and the prediction of protein residue contact.The prediction method is explored and studied at the level of deep learning algorithm model,and an effective corresponding algorithm is proposed Forecasting model.The main work of this study is as follows:(1)This research proposes an end-to-end model that integrates multiple multi-scale convolutions and multi-level bidirectional long-term and short-term memory networks.Compared with other mainstream machine learning-based network models or deep learningbased network models,the two encoding methods are directly encoded.The convolution operation is mixed to extract the sequence feature information of amino acids.This model more fully extracts the local feature information in the two codes.In addition,the model effectively fuses the extracted local short-range feature information and long-range information.Fully mine hidden feature information in protein sequences.The entire algorithm model first performs multiple multi-scale convolutions on the unique hot sequence information of amino acid residues and amino acid evolutionary structure information to extract feature information.The extracted feature information is fused with the original sequence information to form a residual module.The short-term memory network performs local and long-range interactions,and then it is sent to the fully connected network layer to make the final fine-grained level prediction of the secondary structure of the eight types of proteins.And the experimental results show that compared with other benchmark methods,the algorithm model proposed in this research has improved the accuracy of the prediction of the secondary structure of eight types of proteins.(2)Aiming at the problem of low accuracy of remote prediction of protein residue contact,this research proposes a neural network model based on highway's deep-level residual network and attention mechanism.Compared with other mainstream protein residue contact prediction models,This model has the advantages of targeted deep and effective extraction of feature information of protein residue sequences.The entire algorithm model first characterizes the protein amino acid sequence vector and sends it to the deep residual network.Then,the feature information and co-evolution information are two-dimensionally sent to the deep residual network,and then sent to the attention mechanism network.Ingressive regression models predict the probability of contact between any two protein residues.And the experimental results show that compared with other methods,the network algorithm model proposed in this study effectively improves the accuracy of protein residue contact prediction and is a competitive algorithm model.