Robust fine-mapping in the presence of linkage disequilibrium mismatch

What it is

Fine-mapping uses summary statistics from genome-wide association studies (GWAS) together with linkage disequilibrium (LD), the correlation between nearby variants, to identify potential causal variants. The LD usually comes from an external reference panel rather than the study's own participants, and a mismatch between the two is common and can compromise fine-mapping accuracy. The authors developed Robust SparsePro (RSparsePro), a probabilistic graphical model with an efficient variational inference algorithm, to fine-map robustly in the presence of such mismatch. In simulations with varying degrees of mismatch (the journal's main simulation figure uses 3 causal variants), its credible sets had consistently higher power and coverage than existing methods. In fine-mapping cis-protein quantitative trait loci, for 1,353 plasma proteins from the Fenland and deCODE studies, and complex-trait risk loci, its credible sets showed consistently higher enrichment of variants with potential functional impacts and higher replication rates across studies and across LD reference panels. It also identified biologically relevant variants in drug target genes, with plasma PCSK9 as a worked example, and implicated potential regulatory mechanisms for low-density lipoprotein (LDL) cholesterol.