A Design Of Predictor Of Intrinsic Disorder Protein Based On Feature Fusion

With the deepening of the theory of feature fusion, recognition methods based on thefeature fusion are more and more applied, and achieve good recognition effect. The intrinsicdisorder protein (IDP) has been associated with biological function, while lacking structure.In this paper, starting from the feature level fusion, we organize and fuse features of proteinsequences based on the method of kernel canonical correlation analysis and multi-modelfeature fusion, and then realize the structure prediction of intrinsic disorder protein. The mainwork as follows:1. Analysis of intrinsic disorder protein structure predicting. With the application offeatures fusion theory, introducing decision-making level fusion method makes the accuracyof the prediction of intrinsic disorder protein structure have a great advance. Howeverdecision-making level fusion is more inclined to fuse the results of prediction, and not makesgood use of the relationships of characteristics. Decision-making level fusion also makesforecasting information lost. In this paper, the feature level fusion is utilized to predict thestructure of intrinsic disorder protein. This method can make full use of correlations andcomplements of features. The feature level fusion makes further improved predictionaccuracy.2. To fuse features based on kernel canonical correlation analysis. This method can makefull use of correlation of the modal characteristics and discrete degrees of feature vectors tofuse features. According to this, we can make the best use of the composition of amino acidmode, chemical mode of amino acid and physical mode of amino acid. It is helpful to improvethe forecast accuracy.3. A design of predictor of intrinsic disorder protein based on multi-model feature fusionmodel. First, for kernel canonical correlation analysis is based on the correlation, in this papermulti-model feature fusion model is used to organize characters of amino acids. According tothe methods of extraction of characters, amino acids mode can be summed up as follows: thecomposition of amino acid mode, chemical mode of amino acid and physical mode of aminoacid. The three groups of mode include six features, as follows: composition of amino acid,tendency of amino acid, polar of amino acid, hydrophobicity of amino acid, charge of aminoacid and Remark465of amino acid.4. Analyzing forecasting results of the feature fusion predictor. According to the schemeproposed by critical assessment of techniques for protein structure prediction, for a given arbitrary length of protein sequence, we can predict whether the sequence is disorder. Andthen, analyzing the comparison results with predictors of VLXT, VSL2and Adaboost, we canget a better forecast accuracy from the predictor based on KCCA in this paper.