2D similarity kernels for biological sequence classification

Abstract

String kernel-based machine learning methods have yielded great success in practical tasks of structured/sequential data analysis. They often exhibit state-of-the-art performance on tasks such as document topic elucidation, biological sequence classification, or protein superfamily and fold prediction. However, typical string kernel methods rely on analysis of discrete 1D string data (e.g., DNA or amino acid sequences). This work introduces new 2D kernel methods for sequence data in the form of sequences of feature vectors (as in biological sequence profiles, or sequences of individual amino acid physico-chemical descriptors). On three protein sequence classification tasks proposed 2D kernels show significant 15-20% improvements compared to state-of-the-art sequence classification methods.