Extended Data Fig. 9: Enhancer–promoter interactions and conservation of mammalian enhancers in STARR-seq data.

a, A control plot with comparison similar to that shown in Fig. 7a but performed using a set of PWMs that were reversed but not complemented. Note that the enrichment overall is much lower, and that the variance is similar to that observed in Fig. 7a, suggesting that most of the variance in Fig. 7a can be explained by random sampling and not biological effect. b, Conservation of GP5d genomic STARR-seq peaks (orange) and genomic STARR-seq input fragments (blue) measured with average GERP (genomic evolutionary rate profiling) scores. Higher GERP score means higher conservation. Three well-known mammalian enhancers are highlighted (see Methods for details). The average number of base pairs that are more conserved than the average coding base pair in the genomic STARR-seq peaks (170-bp region centered on the peak) was 51 for average peak and 119 for a set of known conserved and biologically important enhancers (MYC335, SOX9 and SHH); this clearly exceeds the ~7 bp conservation expected from the ~15-bit complexity of the TF motifs contained in the active elements derived from random sequence. These results suggest that regulatory elements in vivo are under more complex selection than elements selected for transcriptional activity in our assay.