Machine Learning Based Approaches for Development of Framework for Protein Data Analysis

Abstract

The thesis presents a protein data analytics framework which uses machine newlinelearning methods. It has four contributions which are the part of the framework. In the newlinefirst contribution parallel master slave architecture with text rank methods is used for newlinethe extraction of protein data. Next, in the second contribution finding of essential protein newlineis done by integrating Chi-Square and Support Vector Machine method (Chi-SVM) newlineand also Mean Weighted Average (MWA) and Recursive Feature Elimination (RFE). newlineLater, third contribution helps in identification of protein secondary structure by using newlineproposed Enhanced Fuzzy Random Forest (EFRF) algorithm. Finally, the fourth contribution newlineuses the Metaheuristic Ant Colony Optimization with Path Relinking strategy newline(MACO-PR) and structure knowledge for the identification of protein tertiary structure and its average energy. newlineThe performance and validation of the proposed PDA framework is achieved using newlinedifferent machine learning algorithms and statistical measures such as SVM, Naive newlineBayes (NB), Ant Colony Optimization (ACO), Random Forest (RF), Betweenness Centrality newline(BC), Closeness Centrality (CC), Root Mean Square Deviation (RMSD), Correlation, newlinePredictive Assessment (PPA), Negative Predictive Assessment (NPA), Specificity newlineAssessment (SA) and Sensitivity Assessment (STA). The results show that the newlineproposed PDA framework performs well when compared with the existing methods. newline