基于位置关联权重矩阵及DNA结构信息预测人类剪接位点

Predicting the Splice Sites of Human Genome Based on Position-correlation Weight Matrix and DNA Structural Parameters

人类剪接位点的识别是当前研究的一个重要课题.根据人类剪接位点附近区域的保守性,以位置关联权重矩阵及DNA结构信息作为特征输入参数,应用支持向量机(SVM)对人类基因组中的供体端和受体端剪接位点做了预测.对于供体端,5-fold交叉检验总体预测精度为92.55%,3-way data split检验总体预测精度为92.25%;受体端5-fold交叉检验总体预测精度为90.70%,3-way data split检验总体预测精度为89.87%.

The human splice site recognition is an important problem.The DNA geometric descriptor and position-correlation weight matrix（PCWM）are introduced to describe the conservative segments around spice sites.And the support vector machine（SVM）models combined with the PCWM scoring function and DNA structural features are developed and used to predict the donor and acceptor spice sites of human genome.For five-fold cross-validation,the total prediction accuracies are 92.55% and 90.70% for donors and acceptors respectively.For 3-way data split,the total accuracies are 92.25% and 89.87% for donors and acceptors,respectively.