Automated Classification of Protein Subcellular Location Patterns on Images of Human Reproductive Tissues

Abstract

As one of the most significant characteristics of human cell, subcellular localization plays a critical role for understanding specific functions of mammalian proteins. In this study, we developed a novel computational protocol for predicting protein subcellular locations from microscope cell images in human reproductive tissues. Experiments are performed on a benchmark dataset consisting of 7 major subcellular classes in human reproductive tissues collected from Human Protein Atlas database. We first separated protein and DNA staining in the images with both linear and nonnegative matrix factorization separation methods; then we extracted protein multi-view texture features including wavelet haralick and local binary patterns; finally based on the selected important feature subset achieved by feature selection technique, we constructed ensemble classifier based on support vector machines for predictions. Our experimental results show that 84% accuracy can be achieved through current system, and when only considering the most confident classifications, the accuracy can rise to 98%.