细胞核蛋白质磷酸化位点的预测方法研究

Study on the Prediction Method of Phosphorylation Sites of Cell Nucleus Proteins

以包含细胞核蛋白质磷酸化位点的肽段为对象,将文本向量化方法与机器分类算法相组合,开展了蛋白质磷酸化位点特征筛选和组合模型评价。结果表明:磷酸化位点下游第一个氨基酸及磷酸化位点氨基酸种类是两个重要特征。独热编码与支持向量机的组合模型综合效果最好,它在训练集上的准确率达91.6%,精准度达94.0%,召回率达89.2%。测试集上的结果表明:该模型也显示出了良好的泛化能力。本模型为亚细胞层面磷酸化调控的精细分析提供了有效方法。

In this paper,the text vectorization methods were combined with machine classification algorithms to screen the features of protein phosphorylation site.And the combination models were evaluated for the prediction ability.The result showed that,the first amino acid downstream of phosphorylation site and the type of amino acid at phosphorylation site are two important features.The model of one-hot coding and svm combination showed the best performance with an accuracy of 91.6%,precision of 94.0%and 89.2%recall rate.The model also showed good generalization ability on the test dataset.This prediction model provides an effective approach for the in-depth analysis of phosphorylation regulation at subcellular level.