Protein function prediction based on physiochemical properties and protein granularity

Assigning biological function to uncharacterized proteins is a fundamental problem in the post-genomic age. The increasing availability of large amounts of data on protein sequences has led to the emergence of developing effective computational methods for quickly and accurately predicting their functions. In this work, we extract 353 numerical features from sequences based not only on physiochemical properties but also on protein granularity. A tool in exploratory data analysis, Principal Component Analysis (PCA), is applied to obtain an optimized feature set by excluding poor-performed or redundant features, resulting in 204 remaining features. Then the optimized 204-feature subset is used to predict protein function with k-nearest neighbors algorithm (KNN). This prediction model achieves an overall accurate prediction rate of 84.6%. The experiment results show that our approach is quite efficient to predict functional class of unknown proteins.

I. Introduction

With the rapid growth of genome sequence information, there is increasing concern about using computational methods to predict protein function [1], [2]. Though an astounding number of new proteins have been identified, the functional role of a large fraction of these proteins still remains unknown. As the gap between the amount of protein sequence data and function information, determining protein function at the proteomic scale become one of the main tasks of biological research at post-genomic era. The existing methods of predicting protein function still cannot meet the requirement of technology advancing [3].