Dosage sensitivity of non-coding elements

To maximize power, DEL and DUP calls from the non-redundant combination of the B37 and B38 callsets (as described above) were used for this analysis. Each window was further characterized by its distance to the nearest exon (the minimum distance between any point in the window and any point in the exon) and the pLI score of the gene corresponding to the nearest exon. The pLI score was set to zero for genes with pLI undefined. In the event that exons of 2 genes were equidistant to the window, the max of the two pLI scores was selected.

For a given SV type (DUP or DEL) and a given functional annotation (e.g., VISTA enhancers), each window was characterized by the presence or absence of one or more SV and the presence or absence of one of more genomic features. We observed a depletion of CNV in windows near exons, and in particular near exons of LoF-intolerant genes (see Fig. 5a). As such, we used a Cochran-Mantel-Haenszel estimate of the odds ratios for each SV type/functional annotation, while stratifying for the proximity to the nearest exon as well as that exon’s LOF-intolerance (pLI). Because adjacent windows are not strictly independent observations – i.e., CNV or features may overlap adjacent windows inducing some spatial correlations – we used a block bootstrap method (resampling was performed on blocks of 10 windows) to estimate robust confidence intervals.