Determining The Subcellular Locations Of Proteins Based On Image

Only in correct subcellular location, can protein participate in all kinds of life activities of cells. Functions of a protein will be different in different subcellular location. Therefore, determination of protein subcellualr location is particularly important for determination of protein function. However, the biological experiment methods for subcellular localization of protein are time-consuming, laborious and high-cost. To solve the problem, methods based on experimental data with computation are utilized to locate the subcellular position of protein. However, traditional computational methods of protein subcellular localization based on amino acid sequences are difficult to detect change parts of protein function. Fortunately, computation of protein subcellular localization based on extracted visual information from image can overcome the lack of traditional methods. In recent years, such methods become a hotspot in bioimage informatics.In view of the facts that local binary pattern has a good recognition performance in face recognition, it has been applied to the study of protein subcellular localization. This paper intended to use the protein image datasets of RandTag. We selected the local ternary pattern, noise tolerant local binary pattern, local phase quantization and local configuration pattern as algorithms of feature extraction, then used support vector machine to predict subcellular location of protein, and finally selected the best feature extraction method according to performance of prediction. Moreover, combination of bottom-up algorithm with Shapley value was utilized to filter relevant feature subset of protein subcellular localization according to accuracy of classification with Naive Bayes. Combining LCP with SVM could obtain the best classification accuracy. When using the feature selection based on Shapley value, feature subset of protein image obtained by LCP was the most representative. Then combining the feature subset with classification method of Naive Bayes also achieved the better prediction performance, and its running time was less than that of SVM. The experimental results showed that:LCP was the best feature extraction method for protein subcellular localization. The classification accuracy of the two methods for LCP proposed in this paper was higher than that of SURF.The result of the study can improve the classification accuracy of protein subcellular localization, and is particularly important for protein function research.