Prediction Of Protein Phosphorylation And S-sulfenylation Sites Based On CapsNet And ACNet

Protein is the basic and important part of life body,and protein post-translational modification(PTM)is a main way to regulate protein function,and it is closely related to biological life activities.Therefore,in-depth study of protein post-translational modification is important for understanding proteins with great significance.With the development of biotechnology,the number of protein sequences obtained by people has become extremely large.How to find protein post-translational modification sites from the massive protein sequences is of vital importance to protein research and drug development.The identification of protein posttranslational modification sites through traditional experimental methods often requires a lot of manpower and time,which can no longer meet the current needs.The development of bioinformatics provides solutions for this,many scholars combine machine learning and deep learning to predict the site of protein post-translational modification,which has achieved good results.This thesis uses an integrated neural network based on deep learning to predict phosphorylation and S-sulfenylation modification sites.The main work is as follows:(1)This thesis is mainly based on protein sequence research.Therefore,in the process of data extraction,considering that the protein post-translational modification site is more closely related to the amino acids before and after the modification site,residues of the same length are intercepted before and after the modification site.Then it uses the position-specific scoring matrix,the physical and chemical properties of amino acids,the RECM conversion matrix and the RECM composition feature to encode protein residues,which can express the information of the residues well,and finally uses the information gain(IG)method to perform the feature selection and uses the best features to predict protein phosphorylation and S-sulfenylation sites.(2)This thesis constructs an integrated neural network based on Capsule Network(Caps Net)and Asymmetric Convolutional Network(ACNet)to predict protein phosphorylation and S-sulfenylation modification sites.Caps Net saves the posture information of the target(such as position,rotation,thickness,size,etc.)in the form of vectors,so it can learn very well on small data sets;ACNet uses an asymmetric convolution block(ACB)to replace the standard block convolution,which effectively improves the model's robustness to target flipping and rotation.The framework uses an asymmetric neural network to process sequence features to obtain more informative sequence features,and then applies Caps Net to process the features,making full use of the advantage that the capsule network can express the position of the attribute on the target,that is,it can be perceived the position of each amino acid in the amino acid residues for realizing the prediction of protein phosphorylation and S-sulfenylation modification sites.(3)In order to verify the effectiveness of the network framework in this thesis,four phosphorylation data sets and two S-sulfenylation data sets were compared and analyzed with other network frameworks and existing methods.The AUROC of this method on the four phosphorylation data sets are 0.8880,0.8873,0.8856,and 0.8928,respectively,and the AUROC on the two S-sulfenylation data sets are 0.8839 and 0.8972,respectively.The experimental results show that the network structure in this thesis is better than the commonly used network structures,and the method in this thesis is also better than the existing phosphorylation or Ssulfenylation site prediction methods.(4)Finally,this thesis summarizes the research work of protein phosphorylation and Ssulfenylation prediction,and prospects for future research directions.