Human protein subcellular localization prediction based on error correcting output coding strategy while combining immunohistochemistry image and amino acid sequence

摘要

Human protein subcellular localization prediction is essential for understanding function of proteins. Only when the proteins arrive at the right place at the right time can they perform their functions. In the past decades, various methods based on machine learning have been proposed to predict the subcellular localization of proteins. However, a common drawback is that most of these methods only focus on single signal of proteins, such as amino acid sequences or protein images. In this paper, a novel prediction model based on error correcting output coding (ECOC) strategy was proposed, which named ECOC-locator. It has two advantages, one is that ECOC-locator takes full advantage of amino acid sequence information, for example, rich context information of amino acid sequences, and takes interpretability of protein images into account. The other is that ECOC-locator balances the class contributions of the two signals to the prediction model by employing ECOC. The ensemble classifier integrated 8 classifiers, five of which were trained by immunohistochemistry (IHC) images and the rest trained by amino acid sequence data. The experimental results show that the accuracy of ECOC-locator can reach 40.34%, which fully demonstrates the effectiveness.