Recognition of Human Promoter based on GMM and Rough Set

Identifying Human promoter is the foundation for understanding gene regulation and the key point of large-scale function prediction. Although many algorithms have been proposed, they are rather complex with the performance still limited by low sensitivity and high false positives. In this paper, Gaussian Mixture Model is used to build the model of positional densities of oligonucleotides as the promoter features. Rough Set is applied for analysis the relationship of the binding sites occurence at same time and identify the human promoter sequence. Clustering algorithm which using the inverse of fuzzy likelihood function as the cluster distance is used to estimate the Gaussian Mixture Model optimal number and the parameters of sub-models. And Least Square is used to calculate the mixing proportions. The optimal solution and efficient convergence are obtained. The features for building the predict model are not perfect, so Rough Set Theory is applied for building the promoter information table finding the relationship of the motifs at upstream and downstream of promoter sequence, at the mean time, to predict human promoter sequence. The simulation results show the accuracy of the prediction model is high.