Exploring structural modeling of proteins for kernel-based enzyme discrimination

Abstract

Computational methods play an important role in investigating the relationships between protein structure and function. In this study, we evaluate different graph representations of protein structures for kernel-based protein function prediction. We use shortest path graph kernels and support vector machines to predict whether a protein is an enzyme or not. We present three different and straightforward strategies for modeling protein structures. Accuracy averages for 10-fold cross-validation range from 84.31% to 86.97% for different modeling strategies, outperforming state-of-the-art work.