Fig. 1

Overall workflow of the study. We first identified high-confidence LoFs in genotyping and the imputed data from the BioMe Biobank (a). Then for each gene with a LoF, we partitioned the BioMe individuals into LoF carriers and non-carriers (b) for comparison of 10 CVD-related traits obtained from the Mount Sinai Hospital Electronic Medical Records (MSH-EMR) (c). Next, we performed trait-specific quality-control (QC) by considering that the CVD-traits are affected by certain ongoing medications (d), followed by statistical analyses to robustly identify LoF-genes that were significantly associated with at least one of these CVD traits (e). In the next step (f), we assessed LoF-harboring genes associated with any of the CVD traits by exploring associations between RNA expression levels of these genes and corresponding CVD traits across seven tissues in STARNET [27]. We then selected genes with concordant CVD trait-associations in both BioMe and STARNET (i.e., when LoFs in a gene are associated with low values of a CVD trait, low expression of the same gene is also associated with low values of the trait) (g). For LoF genes associated with lower plasma cholesterol or triglyceride levels in BioMe Biobank and STARNET liver data, we carried out functional in vitro evaluation using HepG2 cells (H). Last, a knowledge-driven filtration approach was used for leveraging information in Gene Ontology (GO) to select potential therapeutic targets for validation in mice (i)