Model-based identification of cis-acting elements from microarray data

Identification of transcriptional regulatory motifs continues to be a challenging problem in computational biology. We report a model-based procedure, MotifModeler, that uses global gene expression data to (1) identify cis-acting elements (CAE) that regulate gene expression under a given condition and (2) estimate the effects of the CAE on gene expression. MotifModeler repeatedly tests random subsets of all possible motifs of a given size and selects those that best fit a combinatorial model of the expression levels. We tested MotifModeler using data from a microarray experiment on the effects of interferon-α in peripheral blood monocytes. Focusing on 6-bp motifs, we predicted 16 stimulatory and 4 inhibitory motifs. Motifs were extended and compared to known binding sites in the TRANSFAC database using position-specific scoring matrices. Many predicted CAE match sites known to be involved in interferon action. MotifModeler demonstrated the potential to computationally identify CAE important in gene regulation.