Research On Prediction Model Of Protein Post-translational Modification Cross-talk

Post-translational modification(PTM)of proteins plays a critical role in various cellular signaling pathways and biological processes.PTMs can regulate signaling pathways by modifying individual residues and by modulating interactions between different modified residues within a protein or between proteins,a phenomenon known as PTM cross-talk.In recent years,computational methods,mainly based on traditional machine learning,have made some progress in predicting PTM cross-talk.Although these methods have addressed the issues of time-consuming and resource-intensive experimental methods,they still face the following problems regarding the accuracy and stability of predicting PTM cross-talk:(1)most computational methods use only protein sequence information or some simple structural information,and there is limited positive data;(2)the instability of prediction accuracy and the limitation of the number of features in computational methods lead to a large number of false positives;(3)current computational methods predict PTM cross-talk separately within proteins and between proteins,and there is no unified method to predict both Intra PTM cross-talk and Inter PTM cross-talk simultaneously.In this thesis,which introduce deeper protein features and establish multiple models to improve the accuracy and stability of predicting PTM cross-talk.The specific content is as follows:(1)To address the problems that previous models for predicting Intra PTM cross-talk only used protein sequence information or some simple structural information when characterizing protein feature information,and lack of positive data led to the dataset imbalanced,an Intra PTM cross-talk prediction model based on multi-feature fusion is proposed,which has added protein structural and dynamic features to the existing features used in previous computational methods to better characterize Intra PTM cross-talk.Additionally,this reaserch have studied the use of imbalanced learning to address the class imbalance in the dataset.Finally,this research have developed models to improve the accuracy of predicting Intra PTM cross-talk.(2)Aiming at the problems of the Inter PTM cross-talk prediction method,such as loss of accuracy,too many false positives,and dataset imbalanced,a model of predicting Inter PTM cross-talk based on multi-model fusion is proposed,which using protein evolution,structural,dynamics features,and imbalanced learning methods,this research found that Inter PTM cross-talk prediction is related to protein-protein interactions(PPIs)and the selfconstructed graph of Inter PTM cross-talk.Based on this,supplementary features such as the graph structure features of PPIs and Inter PTM cross-talk were introduced,then accurate features(Protein evolution,structural,and dynamics features)were combined with generalized features(PPIs and self-constructed graph features).Finally,an heterogeneous network was used to explore the heterogenous relationship between accurate and generalized features,and PPICT model was constructed to predict Inter PTM cross-talk.(3)Due to the lack of a unified model for in-depth mining of these two types of PTM cross-talk and the scarcity of data samples,a model of predicting PTM cross-talk based on multi-layer network is proposed.This research rely on existing models to predict and score hidden Intra/Inter PTM cross-talk,after integration with experimentally verified positive samples,it becomes the initial dataset.Secondly,this research dynamically construct a multi-layer network of the initial dataset and use random-walk methods to learn the multilayer network,thus obtaining the representation features of PTM sites.Finally,this research build the model(MPTC/WMPTC)to predict PTM cross-talk.Through the above three research contents,some problems in the prediction PTM crosstalk task are solved,and the performance of prediction is improved to a certain extent.