【摘要】 目的蛋白质磷酸化是通过激酶催化特定位点把磷酸基转移到底物蛋白质氨基酸残基的过程,是研究蛋白质活力及功能的重要机制。目前已鉴定的数千个磷酸化位点大多缺失激酶信息,为此本研究提出基于PU-learning的磷酸激酶预测算法,通过迭代标记磷酸位点,可以准确预测催化磷酸肽的磷酸激酶。方法首先该算法以PU-learning为框架,利用最大熵方差对不同种类的磷酸激酶自动筛选最佳阈值,从而提取每条磷酸肽上潜在的磷酸化位点,然后根据统计分析确定磷酸化位点对应的激酶,最后通过五折交叉验证该算法在Phospho. ELM数据库上的预测性能,并与现有算法对比。结果该算法的交叉验证特异性和灵敏度比现有最好算法在单个数据集上最高提高4%及10%,其预测Phospho. ELM中数据准确度达到79. 52%。结论基于PU-learning的磷酸激酶预测算法显著优于现有算法,且可以准确预测Phospho. ELM数据库中未知激酶信息的磷酸肽,在磷酸化实验中具有较强的指导意义。更多还原

【Abstract】 Objective Protein phosphorylation is a process by which a kinase catalyzes the transfer of a phosphate group to a protein residue at a specific site,as an important mechanism of protein activity and function. Most of identified phosphorylation sites are lack of kinase information. To this end,a prediction algorithm of phosphokinase based on PU-learning is proposed. By iterative phosphate site labeling,the phosphokinase that catalyzes the phosphopeptide can be accurately predicted. Methods The algorithm uses PUlearning as the framework to automatically screen the optimal thresholds for different kinds of phosphokinases by using the maximum entropy variance,so as to extract the potential phosphorylation sites on each phosphopeptide,and then determines the corresponding phosphorylation sites according to statistical analysis.Finally,the prediction performance is verified by a five-fold cross validation on the Phospho. ELM database and compared with existing algorithms. Results The cross-validation specificity and sensitivity of this algorithm are4% and 10% higher than those of the best existing approach on single data set,and the prediction accuracy on Phospho. ELM is as high as 79. 52%. Conclusions The prediction algorithm of phosphokinase based on PUlearning is significantly better than the existing algorithms,and can accurately predict the phosphopeptides of unknown kinase information in the Phospho. ELM database,which has a strong guiding significance in phosphorylation experiments.更多还原