一种用于模体预测的改进吉布斯采样算法

An Improved Gibbs Sampling Method for Motif Finding

当前有许多用于预测模体的算法,但没有一种算法能有效地应用在所有场合。依据位置权重矩阵的模体模型,提出一种改进的吉布斯采样算法来识别模体。该算法有效地克服了吉布斯采样算法的局部收敛性,并且可以直观地控制预测模体的保守度。同时引入了模体库的概念,并通过分析模体库数据,提高了模体预测的灵活性和准确率。设计了仿真数据,并选择了已被生物实验验证过的模体数据,证实本算法的可行性和有效性。与当前常用的基于吉布斯采样改进的算法比较,本算法有效地提高了模体预测的准确性、灵活性和稳定性。

Many motif-finding programs have been developed, but no program is clearly suitable to in all situations. In this paper, an improved Gibbs sample algorithm was proposed to find motif according to the motif model of Position Weight Matrix （PWM）. The improved approach overcame the local convergence of Gibbs sample algorithm and can control intuitively the conservation for motif finding. The motif base concept was adapted to increase the flexibility and the accuracy for motif finding by analyzing motif data base. The simulated data and the verified biological data are used to test the feasibility and effectiveness of improved approach. Compared with other conventional algorithms, the proposed algorithm increased the accuracy, flexibility and stability effectively for the motif finding.