Research On Protein Palmitoylation Sites Prediction Based On Integrated Deep Neural Network

Palmitoylation of proteins is an important form of post-translational lipid covalent modification and an important mechanism that regulates protein transport,stabilization,localization and function.At the same time,palmitoylation site modification is also involved in a variety of cell biological processes,and is closely related to the occurrence and development of many diseases.The aim of palmitoylation site prediction is to predict whether palmitoylation will occur in a protein and the residues of palmitoylation from the amino acid sequence information and its physicochemical properties.palmitoylation site prediction can help biologists identify the palmitoylation from vast amounts of protein in the data quickly and accurately,reveal the influence about protein folding,activity,and the final function.palmitoylation site prediction can also help ultimately clarify many disease mechanism and overcome certain disease by providing theoretical basis and solutions.Therefore,palmitoylation site prediction has become a hot topic in bioinformatics in recent years.In previous studies,the identification of palmitoylation sites by biological experiments was more reliable.Therefore,the published data of palmitoylated proteins are generally obtained from biological experiments.However,these methods are time-consuming,costly and complicated.In recent years,with the growing of protein database,simple biological experiments have been unable to complete the detection of massive protein data.Machine learning algorithm has been gradually applied in this field,which has greatly improved the detection efficiency.However,there are still some deficiencies such as feature preference and method generalization.Therefore,there is still room for innovation and improvement in the field of protein palmitoylation site prediction.In this paper,an effective method for prediction of palmitoylation sites was established by combining multi-angle features with neural network to solve the problems of single feature,limited model and unsuitability for large amount of complex data processing.This method selects four feature sets from two perspectives,constructs four complex submodules of deep convolutional neural network according to different feature dimensions,and integrates the processing results of independent feature information channels to predict their sites.In the test experiment,the method calculates the corresponding evaluation value from the whole model and the multi-feature level respectively.The overall experimental results show that compared with SeqPalm,a popular prediction tool based on traditional machine learning algorithm,the multiple evaluation indexes measured by this method are more balanced,showing better stability and generalization performance.Compared with the migration model CapsNet,the accuracy and sensitivity of the prediction results in this experiment were improved.In the multi-feature experiment,compared with the single-feature model,the ROC curve of the comprehensive feature model is smoother and the model is more stable.Based on the above experimental results,this method improves the accuracy and reliability of site prediction compared with the existing palmitoylation site prediction methods,which proves the potential capability of this method in its research field.Some experimental steps of the method used in this paper are slightly complicated,and there is still some room for improvement.