Abstract
BackgroundClassification of protein sequences is a central problem in computational biology. Currently, among computational methods discriminative kernel-based approaches provide the most accurate results. However, kernel-based methods often lack an interpretable model for analysis of discriminative sequence features, and predictions on new sequences usually are computationally expensive.ResultsIn this work we present a novel kernel for protein sequences based on average word similarity between two sequences. We show that this kernel gives rise to a feature space that allows analysis of discriminative features and fast classification of new sequences. We demonstrate the performance of our approach on a widely-used benchmark setup for protein remote homology detection.ConclusionOur word correlation approach provides highly competitive performance as compared with state-of-the-art methods for protein remote homology detection. The learned model is interpretable in terms of biologically meaningful features. In particular, analysis of discriminative words allows the identification of characteristic regions in biological sequences. Because of its high computational efficiency, our method can be applied to ranking of potential homologs in large databases.
Highlights
Classification of protein sequences is a central problem in computational biology
We provide some further analysis of the associated sequence representation, which gives rise to a well interpretable feature space in terms of "word correlation matrices" (WCMs)
We presented a new approach for protein sequence representation based on word correlation matrices (WCM)
Summary
Classification of protein sequences is a central problem in computational biology. Currently, among computational methods discriminative kernel-based approaches provide the most accurate results. I.e. sequences with a similarity of more than 80% at the amino acid level, this can be done by pairwise comparison methods like the Smith-Waterman local alignment algorithm [1] or BLAST [2]. These methods often fail in cases where sequence similarity is low. Remote homology detection methods are often based on a statistical representation of protein families and can be divided into two major categories: first, profile-based methods provide a non-discriminative approach to family-specific representation of sequence properties.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
