When Less Is More: Improving Classification of Protein Families with a Minimal Set of Global Features

Abstract

Sequence-derived structural and physicochemical features have been used to develop models for predicting protein families. Here, we test the hypothesis that high-level functional groups of proteins may be classified by a very small set of global features directly extracted from sequence alone. To test this, we represent each protein using a small number of normalized global sequence features and classify them into functional groups, using support vector machines (SVM). Furthermore, the contribution of specific subsets of features to the classification quality is thoroughly investigated. The representation of proteins using global features provides effective information for protein family classification, with comparable results to those obtained by representation using local sequence alignment scores. Furthermore, a combination of global and local sequence features significantly improves classification performance.