Prediction of Linear B-cell Epitopes Based on PCA and RNN Network

Epitope prediction plays an important role in production of antibodies and disease treatment. There are mainly two research methods, namely experimental method and calculation method. Experimental method can obtain more accurate experimental results, but it takes a long time and the cost of manpower, material resources are relatively high. So it is not convenient to obtain experimental results more quickly. Calculation method mostly uses computer and machine learning methods for prediction. Calculation method improves prediction speed to some extent, but the result is not satisfactory. In order to further improve the accuracy of epitope prediction, this paper proposes a novel method of processing epitope characteristics. In this paper, we choose six properties to study. The six main physicochemical properties are converted into corresponding digital vectors, resulting in high-dimensional features. Then we use Principal Component Analysis (PCA) method to process them. Finally, dimensionality reduction features are used as input of Recurrent Neural Network (RNN) for epitope prediction, and good prediction results are obtained. PCA method reduces feature dimensions and facilitates the processing of features. At the same time, the prediction results obtained with dimensionality reduction features show that dimensionality reduction reduces dimensions, but it retains the main components of original features and improves the rate of successful prediction.