Fig. 5

De novo sequence motif prediction. a Scheme for de novo motif prediction. The sequences of 3′ UTR were cut into 100 bases long bins with 50 bases overlap in between adjacent bins (Step 1). From each gene one bin is randomly selected and sets of bins from each gene were randomly generated for hundreds of times (Step 2). De novo motif prediction was performed over the sets (Step 3). Motifs found are then subjected to over-representation analysis (Step 4). b The resulting over-represented motifs in Cluster I and (c) their sequence logos. d Number of hits in quick, slow and stable gene sets. Hypergeometric p values with Bonferroni correction are also shown. Red letters represent significant (p < 0.01) differences