Protein Subcellular Localization Prediction by Concatenation of Convolutional Blocks for Deep Features Extraction From Microscopic Images

Understanding where proteins are located within the cells is essential for proteomics research. Knowledge of protein subcellular location aids in early disease detection and drug targeting treatments. Incorrect localization of proteins can interfere with the functioning of cells and leads to illnesses like cancer. Technological advances have enabled computational methods to detect protein's subcellular location in living organisms. The advent of high-quality microscopy has led to the development of image-based prediction algorithms for protein subcellular localization. Confocal microscopy, which is used by the Human Protein Atlas (HPA), is a great tool for locating proteins. HPA database comprises millions of images which have been procured using confocal microscopy and are annotated with single as well as multi-labels. However, the multi-instance nature of the classification task and the low quality of the images make image-based prediction an extremely difficult problem. There are probably just a few algorithms for automatically predicting protein localization, and most of them are limited to single-label classification. Therefore, it is important to develop a satisfactory automatic multi-label HPA recognition system. The aim of this research is to design a model based on deep learning for automatic recognition system for classifying multi-label HPA. Specifically, a novel Convolutional Neural Network design for classifying protein distribution across 28 subcellular compartments has been presented in this paper. Extensive experiments have been done on the proposed model to achieve the best results for multilabel classification. With the proposed CNN framework as F1-score of 0.77 was achieved which outperformed the latest approaches.