A Comparative Study between Fixed-size Kernel Logistic Regression and Support Vector Machines Methods for beta-turns Prediction in Protein

Murtada Khalafallah Elbashir


Beta-turn is an important element of protein structure; it plays a significant role in protein configuration and function. There are several methods developed for prediction of beta-turns from protein sequences. The best methods are based on Neural Networks (NNs) or Support Vector Machines (SVMs). Although Kernel Logistic Regression (KLR) is a powerful classification technique that has been applied successfully in many classification problems, however it is often not found in beta-turns classification, mainly because it is computationally expensive. Fixed-Size Kernel Logistic Regression (FS-KLR) is a fast and accurate approximate implementation of KLR for large-scale data sets. It uses trust-region Newton’s method for large-scale Logistic Regression (LR) as a basis, to solve the approximate problem, and Nystrom method to approximate the features' matrix. In this paper we used FS-KLR for beta-turns prediction and the results obtained are compared to those obtained with SVM. Secondary structure information and Position Specific Scoring Matrices (PSSMs) are utilized as input features. The performance achieved using FS-KLR is found to be comparable to that of SVM method. FS-KLR has an advantage of yielding probabilistic outputs directly and its extension to the multi-class case is well-defined. In addition its evaluation time is less than that of SVM method.

Full Text: PDF


  • There are currently no refbacks.