Figure 3

Overview of motif extraction pipeline. Schematic overview of motif processing steps. Gene structure annotations are projected across the whole genome alignments. Motif candidates are identified on a subset of 9,500 randomly picked upstream regions. Degenerate motif descriptions are removed if the set of atomic motifs, which they represent, scores better in terms of conservation enrichment. The greatest reduction in the number of candidate motifs is attained by scoring conservation (Z-Score and P-value filter with a 5% FDR level cutoff). Additionally, larger motifs are removed if smaller substrings (≥ 6 bp) of these motifs score better in terms of conservation. Motif candidates are then evaluated by a non-parametric test, which assesses their influence on gene expression. Finally, conditional trees are employed to select motif ensembles, which possibly have a joint regulatory function.