Research Of Protein Ubiquitination Site Prediction Method Using Deep Learning

Protein ubiquitination modification is one of the important post-translational modifications of proteins,which plays a very important role in apoptosis,transcriptional regulation,cell disease,DNA repair and other basic reactions.Efficient and accurate identification of ubiquitin sites is of great biological significance for the study of protein ubiquitin modification.Traditional biometric-based methods require researchers to carry out biological experiments in a large amount of protein sequence dataset,such as CHIP-CHIP analysis and mass spectrometry,which requires a lot of time and economic cost.The computing-based recognition method can predict ubiquitylation sites efficiently and accurately on large-scale data,and most computing-based methods only need protein sequence information,so it is easier to identify,therefore,the computing-based method arises at the historic moment.In this thesis,the ubiquitination site sequence dataset is obtained from the protein lysine modification database,and the sequence is preprocessed using bioinformatics tools to obtain a positive and negative sample set.Combine one-of-key coding and amino acid physical and chemical properties to encode sequence data.A prediction model based on deep learning is designed.The model mainly consists of two modules,namely the attention module and the capsule network module.The attention module can selectively focus on the importance of amino acids in the sequence,and the capsule network module can reflect the spatial position relationship of the internal characteristics of the neural network.Therefore,the capsule network model combined with the attention mechanism in this thesis can encode the dependency between protein sequences and capture important amino acid features in the sequence.Experimental results show that the accuracy,specificity,sensitivity,AUC,and MCC values of this model are better than other existing prediction models in most cases.