Voting for the Prediction of Protein Secondary Structure and Its Evaluation

Protein secondary structure prediction is one of the central topics in proteome analysis. Computational methods, developed for the prediction (classification) of protein secondary structures, have been improved substantially since 1990s, allowing us to investigate some of the computational classifiers and attempt to integrate them through voting. The study tries to evaluate whether and how much voting can improve the prediction accuracy. In the research, 4 classifiers (i.e. predictors), SSpro, PSIpred, PHD and Prof, are selected since they produce some reasonably good prediction accuracies. Two voting methods are adopted to integrate these 4 classifiers – a simple majority voting by assigning data to a class that gains the majority votes, and a weighted majority voting which weights each vote by the prediction accuracy of the classifiers. The voting results show that including better-performed classifiers tends to improve the prediction while including poor-performed classifiers tend to deteriorate the prediction. More investigation could be carried out using more classifiers or more diverse classifiers in a future research.