The human proteome harbors approximately 1500 TFs [43], although TFBS models are available through HOCOMOCO for only 401 TFs. To compensate for this and to allow the detection of TFBSs whose motifs remain unknown, we applied ab initio motif discovery to genome-wide promoters, in order to complement the HOCOMOCO results. Ab initio identified motif families (MFs) generated by the Dragon Motif Finder [44], suggest multiple levels of sequence complexity specific to lncRNA promoters. These include reverse-complement motifs (palindromes) unique to lncRNA promoters, long motifs (20 bps), and polyA/polyT-rich regions (Figure S2a–d).
