Research On Protein Secondary Structure Prediction Based On Multiple Classifiers Combination

As a basic research in the field of bioinformatics,the prediction of the protein secondary structure is of great significance in determining the spatial structure of protein and clarifying the function of protein.Although the protein structure can be measured through experiments,it is difficult to meet the increasing demand of protein sequence data processing,so it is imperative to predict protein structure by means of machine learning.Because the existing single-classifier secondary structure prediction method is difficult to improve the effectiveness,in this paper,a protein secondary structure prediction method based on multi-classifier combination is used as the research object.The main contents are as follows:Firstly,the molecular composition information,structure classification information and common protein database of proteins are summarized,and the homomorphic combination learning and heteromorphic combination learning are discussed.The multi-classifier combination learning method is carefully summarized and reviewed.As the construction of the feature vector and the design of multi-classifier combination algorithm are two important sections of protein secondary structure prediction method based on multi-classifier fusion,the above review provides the theoretical basis and application premise for the research of this paper.Secondly,in this paper,a method for prediction of protein secondary structure based on weighted combination with multiple evolutionary matrices is proposed.In this method,the score matrix based on near-correlation protein alignment and the score matrix based on distant correlation protein alignment are used as input vectors of support vector machine(SVM).The posteriori probability information of member classifier output is processed by weighted combination method,and the weight of each classifier is decided with the individual classifier's error on the training set,and a multi-classifier model based on weighted combination method is constructed.The experimental results show that this method can effectively improve the accuracy of the protein secondary structure prediction.Finally,this paper proposes a dynamic adaptive weighted combination protein secondary structure prediction method based on entropy.In this method,two kinds of weighting coefficients are designed.One is to adjust the weight according to the entropy calculated from the information of the sample posteriori probability output by the member classifier.The greater the entropy value is,the lower the combination weight is given by the classifier with higher entropy value.The second is to dynamically adjust the weighted parameters according to the "confidence" degree of classification results.Finally,the combination is realized by weighted voting,and the final prediction results are obtained.The experimental results show that this method can effectively improve the accuracy of protein secondary structure prediction.