Fig. 1

Study overview. a Overview of the three PBMC scRNA-seq studies used in our study. The studies, the version of the used chemistry for data generation (version 2 and 3, referred to as v2 and v3, respectively), number of individuals involved (indicated as the number in the parenthesis), and relative composition of the major blood cell types used in this study. b Co-expression benchmarking scheme. We first benchmarked co-expression patterns among the three scRNA-seq studies and compared them to co-expression patterns in different bulk datasets. As an additional external validation, we benchmarked both the scRNA-seq and bulk co-expression patterns with a CRISPR knockout dataset. After benchmarking, we evaluated differences in co-expression patterns among cell types and among individuals within a cell type. c Co-expression QTL (co-eQTL) mapping. Building on the benchmarked co-expression pattern, we developed a novel strategy to identify co-eQTLs (genetic variants changing co-expression). Part of the strategy is a strict filtering of tested SNP–eGene–co-eGene triplets, where the SNP is required to be an eQTL for one of the genes and the genes show significant correlation in at least a certain number of individuals. d Co-expression QTL (co-eQTL) insights. We first check if the SNP we tested has been identified in previous GWAS studies, or it is in high LD with GWAS variants. Then we check the group of co-eGenes on what pathways and traits and TF binding sites they are enriched for. We hypothesize two possible scenarios of the underlying biological mechanism for the identified co-eQTLs. Scenario 1 portrays a direct regulatory association between the co-eGene and the eGene through the genetic variants that change the binding affinity. Scenario 2 portrays an indirect association between the co-eGene and the eGene. That is, the co-eGene is co-expressed with the TF that regulates the expression level of the eGene