基于AdaBoost方法的蛋白质磷酸化修饰规则抽取

Protein phosphorylation rules extraction based on AdaBoost

磷酸化是最重要的蛋白质翻译后修饰之一,随着蛋白质磷酸化数据的增加,利用已有数据对蛋白质磷酸化修饰进行规律挖掘和预测的条件日益成熟。设计新的基于AdaBoost(adaptivc boost)分类器的规则抽取算法和利用修饰位点附近氨基酸性质作为特征并采用AdaBoost方法进行特征选择和分类器训练的磷酸化修饰位点预测方法AproPhos(using amino acid pro- perties for phosphorylation sites prediction),使其在具有较高预测精度的同时可以对预测结果进行可理解的规则解释,规则抽取还有助于发现新的磷酸化修饰氨基酸性质分布规律,对揭示生命活动规律和药物开发有着重要意义。

Protein phosphorylation is one ofthe most important post-translational modifications. With the recent increase in protein phosphorylation sites identified, phosphorylation rules mining and potential phosphorylation sites prediction may facilitate the research of phosphorylated protein. A new algorithm for rule extraction from AdaBoost （adaptive boost） and a new phosphorylation sites prediction method named AproPhos （using amino acid properties for phosphorylation sites prediction） using AdaBoost as amino acid properties feature selection and classification are designed. They can provide understandable explanation of the prediction at the same time they perform higher prediction accuracy. Rule extraction may be helpful to discover new rules of amino acid properties distribute around sites and be significant for the research of life sciences and medicine development.