Research Of Protein Subcellular Localization Prediction Based Deep Learning

High-content screen is used in genetic analysis and environmentally disturbed gene mutation in the research of cell biology.The emergence of High Content Screening,makes the biologists be able to design the experiments about inferring relationships between different gene mutation and cell morphology at different growth cycles.Despite the benefits of experiments based on High-content screening technique,analyzing a whole bunch of High-content images remains a challenging to us.High-content images contain lots of information and features so that how to extract the information we want becomes one of the most difficult problems,such as protein subcellular localization.Some research groups still predict the protein subcellular localization of cells inside the High-content images manually by eye.Other groups have developed computational analysis to predict protein subcellular localization based on conventional machine learning algorithm.In this paper,I apply a computational analysis method about protein subcellular localization prediction in the High-content images based on deep learning.I design a convolutional neural network,display the performance of the deep learning based method and the machine learning based method in the fifteen labels classification task on the datasets of fluorescently tagged proteins in yeast cells.Meanwhile,I show the reasons why the convolutional neural network reach the performance through the features distribution,features visualization and transfer learning.I prove that in such a classification task our algorithm outperforms the conventional machine learning based method and the convolutional neural network is able to extract highlevel features.I further prove that our model can be used in different classification task,including different labels in the same dataset as well as different dataset.At last I use the model to analyze nearly 200 proteins that change in localization in response to mating pheromone.