A hybrid approach of support vector machines with logistic regression for β-turn prediction

A β-turn is a secondary protein structure type that plays a significant role in protein folding, stability, and molecular recognition. It is the most common type of non-repetitive structures. On average 25% of amino acids in protein structures are located in β-turns. In this paper, we propose a hybrid approach of support vector machines (SVMs) with logistic regression (LR) for β-turn prediction. In this hybrid approach, the non β-turn class in a training set is under-sampled several times and combined with the β-turn class to create a number of balanced sets. Each balanced set is used for training one SVM at a time. The results of the SVMs are aggregated by using a logistic regression model. By adopting this hybrid approach, we cannot only avoid the difficulty of imbalanced data, but also have outputs with probability, and less ambiguous than combining SVM with other methods such as voting. Our simulation studies on BT426, and other datasets show that this hybrid approach achieves favorable performance in predicting β-turns as measured by the Matthew correlation coefficient (MCC) when compared with other competing methods.