Research On Intrinsically Disordered Proteins Prediction Through Machine Learning Methods

The looseness of intrinsically disordered protein's own structure makes it able to bind with a variety of biological macro-molecules,occupies an important role in cell function regulation and signal transduction,and has a close relationship with many major human diseases,and has become a hot spot in current research.However,due to the looseness of the disordered protein's own structure,it is impossible to form a stable three-dimensional structure in the natural state,making it difficult to determine it through biological experiments.Prediction through computational methods becomes an effective approach in the study of disordered proteins.For this reason,this paper introduces deep learning algorithm and compares it with traditional machine learning algorithms to provide an effective approach for the research of predictive methods for intrinsically disordered proteins.This dissertation firstly constructs data sets and extracts eigenvalues based on Disprot's disordered protein database.Based on this,the evolutionary conservative matrix PSSM and the physicochemical properties of amino acids are selected by optimizing the combination of feature sets used by the existing prediction algorithms.With the three characteristics of amino acid composition,an combined feature coding model was constructed.By using traditional machine learning algorithms(logistic regression,linear discriminant analysis,K nearest neighbors,naive Bayes,decision trees and support vector machines)and deep learning algorithms(convolutional neural networks and recurrent neural networks)in disordered proteins' prediction,the application of the prediction algorithm was compared and analyzed.Different data sets were designed for predictive experiments of disordered proteins and different combinations of feature data sets were used for prediction experiments.The results showed that the accuracy of the deep learning algorithm was significantly better than that of traditional machine learning methods.The learning algorithm has better efficiency than the support vector machine and can be applied to the classification prediction of disordered proteins.At last,based on machine learning algorithms including deep learning methods,the design and implementation of the intrinsically disordered protein prediction system are given,and corresponding services are provided for the further research of the intrinsically disordered protein.