Protein Subcellular Localization Model Based On Deep Learning

Protein is the performer of life activities,and the important structural component of an organism.In biological cells,proteins are participating in many life activities and play many important functions.The exercise of normal functions by proteins is subject to the biochemical environment in which they are located.Abnormal protein localization is often associated with diseases,such as Alzheimer's disease,metabolic disorders and cancers.Therefore,protein subcellular localization studies play a key role in understanding protein functions and mechanisms and are of great importance in biology and medicine.The detection of protein localization by experimental means requires too much labor and time to satisfy people's needs,which makes it possible for computational methods to play an important role in this area.Current computational methods can be broadly classified into three categories.The first category is homology-based methods,which find highly homologous sequences of a target sequence from databases and make predictions based on the subcellular localization information of the homologous sequences.The second category of methods is machine learning methods,the principle of which is to train models from collected training data based on sequence information and manually extracted features for protein subcellular localization prediction.In the third category,deep learning methods have powerful learning capabilities and are widely used for subcellular localization prediction tasks with good experimental results.The shortcoming of deep learning methods is that they only make use of the sequence information of the protein and do not make use of the structural information of the protein.In the spatial structure of a protein,neighboring amino acids affect each other and thus affect the properties of the protein.Therefore,it is necessary to make full use of protein structure information and capture the correlation between protein structure information and protein localization.This paper presents Graph Loc,a protein subcellular localization prediction model based on deep learning methods,with the main innovation being the application of graph neural networks in the subcellular localization prediction task.In this paper,the predicted protein contact graph is used to represent the spatial structure of a protein,and the protein structure information is effectively utilized by aggregating the information of adjacent amino acid features on the space through a graph convolutional network,and a self-attention mechanism is used to assign weights to the amino acids in the protein,balancing the sequence information and the structural information of the protein.The experiments were performed on more than 12,000 proteins on the Deep Loc dataset.The experimental results show that our model achieves the highest accuracy.In the comparison by category,our model outperformed other current algorithms on most of the metrics,indicating that the method has potential implications in protein subcellular localization prediction.The method may be applied to protein localizationrelated disease research in the future.