Comparison and analysis of models predicting transcriptional regulatory modules based on different backgrounds

Correct recognition of transcriptional regulatory elements (also named motif) is important for understanding the laws of expression of genes. In silicon analysis, generally, a background or named control set constructed by a set of sequences is necessary in predicting transcriptional regulatory elements. Some studies have suggested that the accuracy of models could be improved when selecting backgrounds according to GC-contents. For further examine control set´s influence on models predicting transcriptional regulatory modules, 3 different kinds of transcriptional regulatory element-recognizing control sets, which are a background from given sequences, a background from shuffled sequences and a background from Markov model, are introduced. Then comparison and analysis of module-predicting methods based on the above 3 kinds of control sets are performed. The results suggested that the better accuracy of prediction is obtained when using a background from Markov model which considers the composition bias of the nucleotides in the biological sequences, while the accuracy of models would be significantly improved when combining different backgrounds.