Research On IDPs Prediction Based On Multi-scales And Multi-features

Intrinsically Disordered Proteins(IDPs)opens a new era for proteins.IDPs can carry out many functions based on a flexible and plastic disordered state.The existence of IDPs breaks the traditional structural level system of protein.Thus the research on IDPs has great revolutionary significances.Although many important IDP-related achievements have been obtained,this researching field is still experiencing,such as the less number of proteins with experimentally determined disordered structure and the unclear mechanisms of disordered structure formation.The predictors for IDPs become one of effective tools to solve these problems.Here we are focus on the method of IDPs predicting modeling,including the studies of feature selection and predicting model.The main contributions are as follows:(1)A method is proposed to flexibly choose the window sizes in the prediction of IDPs.Because the existing methods for the selection of window sizes can be insufficient to meet the requirements for the multi-scale characteristics of IDPs and have a high subjective content,a dynamic method is proposed.Taking account of the multi-scale characteristics of IDPs,amino acid sequences is observed based on many different window sizes.In order to make an objective selection,the diversity among these predictors based on these different window sizes is estimated and then the window sizes can be chosen according to estimated results.(2)The bias levels of amino acid to different structure are analysized and a new attribute called structural bias level is proposed.The folding mechanism of structure formation of proteins is believed to conceal in amino acid sequences.The structural bias of amino acids is believed as the expression of the concealed information.In order to extracting the more information of IDPs,structural bias levels of amino acids is considered as new parameters to express the characteristics of IDPs.The ability of amino acids to forming disordered structure is quantitatively estimated by used of the probability knowledge and a score form is constructed.The results of predictors based on structural bias level shows that the information from structural bias is effective to improve the predicting accuracy of IDPs.(3)Predicting teams is constructed for IDPs.Due to the unclear folding mechanism of structure forming and function foundation of IDPs,there are many uncertainties between extracting information and disorder predicting results.In order to provide the more information,the predicting ability of each attribute to different disordered structure is computed by measuring the mutual Information.For different disordered structure,different predicting teams are constructed which include the stronger predicting ability to corresponding disordered structure.Predictors are built based on these predicting teams and the predicting results show predicting teams are good to the prediction of IDPs.(4)The predicting model for IDPs based on different scales and different features are built.Take account of the complex process for folding structure of IDPs,this problem can be divided into some simply problems.That is,many predictors are built based on different window sizes and features for different disordered structure.By measuring diversity among these predictors,some with larger diversity are selected.Making use of majority vote rule,the predicting model for IDPs based on multi-scales and multi-features is built.