MLEP:一种B细胞线性表位预测方法

为了更快更准地确定B细胞线性表位,提出了一种新的预测方法——MLEP(Prediction of epitope based on MCFS and LSTM,MLEP)算法。采用5种性质氨基酸理化性质作为学习特征,利用多聚类特征选择算法进行特征选择,用降维后的数据作为输入,用长短期记忆网络进行训练,获得预测性能好的模型,对多聚类特征选择算法及MLEP算法的性能进行评价。对非冗余LBtope数据集进行多组实验,结果表明,使用多聚类特征选择算法降维到25时获取性能最优模型,多聚类特征选择算法比主成分分析法获得的模型准确率更高,基于MLEP算法获得的模型准确率达到94.81%。因此,MLEP算法能更好地预测B细胞线性表位,对于表位预测研究具有一定的参考价值。

MLEP: A new method for prediction of linear B-cell epitopes

In order to determine the linear B-cell epitope faster and more accurately, a new prediction method MLEP algorithm is provided. Firstly, all the prediction calculations are based on the five properties scales of amino acids. Based on these results, a multi-cluster feature selection algorithm is studied for reducing the number of dimensions. Secondly, the networks is trained using long-short term memory network algorithm and with the reduced dimension data. Finally, the performance of the multi-cluster feature selection algorithm and the MLEP algorithm is evaluated. The experimental evaluation of classification is performed using the non-redundant LBtope dataset. The results show that the multi-cluster feature selection algorithm achieves the best performance when the dimension is reduced to 25, and the performance of the multi-cluster feature selection algorithm is significantly better than the methods based on the principal component analysis, and the maximum accuracy of 94.81% can be achieved using the MLEP algorithm. This method can effectively predict the linear epitope of B cells, which provides reference for the study of epitope prediction.