Fig. 6: Conservation and enrichment of genomic variants in PrismNet-predicted HARs.
From: Predicting dynamic cellular protein–RNA interactions by deep learning using in vivo RNA structures
a Distribution of PhyloP20way conservation scores of three groups of HARs (i.e., Sequence only, Sequence & Structure, and Structure only), with conservation scores for all CLIP binding sites and randomly sampled transcript regions as positive and negative controls, respectively. ***P < 0.001, one-sided unpaired t-test. b Enrichment of common SNPs (from dbSNP) in three groups of HARs and CLIP binding sites for every RBP, compared against randomly sampled transcript regions. ***P < 0.001, one-sided paired t-test. c Enrichment of rare SNVs (from dbSNP, Minor allele frequency (MAF) < 0.05%) relative to common SNVs in three groups of HARs and CLIP binding sites for every RBP. RR, randomly sampled transcript regions. *P < 0.05, **P < 0.01, ***P < 0.001, one-sided paired t-test. d Enrichment of ClinVar mutations in HARs, compared to randomly sampled transcript regions. ***P < 0.001, permutation test. e, f Enrichment of ASD, BD, and SCZ-associated variants among all HARs collectively (e) or HARs of every RBP respectively (f), compared to randomly sampled transcript regions. RBPs previously reported to be associated with psychiatric disorders are highlighted. ***P < 0.001, permutation test. The data of ASD_1 are from An et al., 2018;77 the data of ASD_2, BD and SCZ are from Gandal et al., 2018.76 All HARs and CLIP-seq binding sites were here defined using data from K562 cells. In Fig. 6b, c, f, each dot represents an RBP. RBPs with HARs significantly depleted/enriched for genetic variants (including common and rare SNVs and disease-associated variants) are shown in color. Gray dots represent the non-significant RBPs. ***P < 0.001, permutation test for Fig. 6b, f, Fisher’s exact test for Fig. 6c.
