Constrained Nonnegative Matrix Factorization for Image-based Protein Subcellular Localization Prediction

Abstract

Protein subcellular location is an important biological information for understanding protein’s function in normal cells. Automatic analysis of protein subcellular location based on bioimage has been received much attention in recent years. Since preprocessing is a critical step in the automatic image-based analysis system for source separation, this research focuses on the protein subcellular location. Some problems exist in most existing separation methods, such as, the lack of strong explanation and low accuracy. In this paper, a new separation method called minimum volume constrain nonnegative matrix factorization for image preprocessing has been proposed. To examine the effectiveness of the proposed method, both local and global features are extracted from the separated channels, and multi-label classifier is used to make prediction for subcellular localization. The results show the proposed method can generally improve the accuracy of final prediction compared with other methods.