Protein subcelluar localisation prediction with improved performance

Abstract

Predicting the subcellular localisation of proteins is crucial for the determination of protein functions. In this paper, we present a computational method for protein localisation prediction. We start with a simple approach that predicts protein localisation based on Euclidian distance computed from residue composition. Then the performance is gradually improved by introducing a weighted Euclidian distance, including homologous information, and using feature selection. The final method achieves 90.3% accuracy in assigning proteins into five subcellular locations. Comparisons with CELLO, PSORT-B and P-CLASSIFIER show that our method outperforms the others.