Fig 1.
(A) Overall architecture of MetaChrom. The input sequence is fed into both MetaFeat and the ResNet sequence encoder. Their outputs are then concatenated for the prediction of epigenomic profiles.(B) Pipeline for predicting variant effect on sequence epigenomic profiles.
Fig 2.
(A) AUROC and (B) AUPRC performance comparison of MetaChrom and DeepSEA method across 31 epigenomic features. NPC, Glut, DN, GA: iPSC-derived neurons. GZ, CP: germinal zone and cortical plate. OCR: open chromatin regions. See Table A in S1 Table for the list of cell/tissue types.
Fig 3.
Validation of MetaChrom predicted functional variants with evolutionary constraint.
(A) Distribution of GERP scores between MetaChrom predicted functional variants and random variants. (B) Minor allele frequencies of variants defined by MetaChrom scores in four selected cell types. Only variants inside peak regions of the epigenomic data were considered.
Fig 4.
Validation of MetaChrom predicted functional variants with ASC variants.
(A) Enrichment of ASC variants for predicted functional variants identified by MetaChrom, Funsig, CADD, deltaSVM, baseline CNN score in iN-Glut and NPC cells. (B) The observed allelic imbalance vs. MetaChrom predicted effects on chromatin accessibility of ASC variants in Glut neurons. (C) Accuracy of predicting directions of ASC variants in Glut and NPC cells.
Fig 5.
MetaChrom score of 31 candidate SNPs across 31 cell types.
Candidate SNPs are ordered by their posterior inclusion probability (PIP) values shown in the middle. Three columns on the right indicate if a SNP is an eQTL (+/-), if a SNP has HiC targets (+/-) and if a SNP acts mostly in fetal stage (F) or in adult stage (A) or both (FA).
Fig 6.
Likely causal variant rs2304205 and its MetaChrom functional annotations.
The candidate SNP rs2304205 is chose as the reference variant for computing LD and it is highlighted by the red dash line in each panel. The upper panel shows significance of GWAS SNPs, LD between SNPs and genes in this region. The next panel shows credible set SNPs identified by fine-mapping (PIPs) in this region. The remaining panels show MetaChrom scores in four cell types, two in fetal stage (FB OCR and Glut) and two in adult stage (VLPEC neuron and PUT neuron).
Fig 7.
(A)Distribution of matched motifs in each epigenomic assay and (B)selected binding motifs identified by our method and their matches in the CIS-BP database.