An improved method to enhance protein structural class prediction using their secondary structure sequences and genetic algorithm

Many approaches have been proposed to enhance the accuracy of protein structural class. However, such approaches did not cover the low-similarity sequences which are proved to be quite challenging. In this study, a 71-dimensional integrated feature vector is extracted from the predicted secondary structure and hydropathy sequence using newly devised strategies for the purpose of categorising proteins into their major structural classes: all-α, all-β, α/β and α+β. A new combined method containing two machine learning algorithms has been proposed for feature selections in this study. Support vector machine (SVM) and genetic algorithm (GA) are combined using the wrapper method for the purpose of selecting top N features based on the level of their importance. The proposed method is evaluated using the jackknife upon two low-similarity sequences datasets, i.e. ASTRAL and D640. The overall accuracies of 83.93 and 92.2% are reported for the predictions pertaining to ASTRALtesting and D640 benchmarks, exceeding most of the current approaches.