Fig 1.
Comparison of methods in simulated data with the in-sample LD matrix.
We assess the 95% credible sets generated by five fine-mapping methods (SuSiE, SuSiE2, mvSuSiE, flashfm, fastPAINTOR) under two scenarios (a) (A, B, C) and (b) (D, E, F). The results, averaged over 100 repetitions for each method and heritability combination, are presented with both the mean value and the empirical standard error. Panels A and D evaluate the power of detecting causal SNPs in at least one credible set. Panels B and E focus on the coverage of credible sets, with the black dashed line indicating the 95% level. Panels C and F evaluate the average size of credible sets for scenarios (a) and (b), respectively.
Fig 2.
Comparison of methods in simulated data with the 1KG reference panel.
We compare the 95% credible sets from five fine-mapping methods (SuSiE, SuSiE2, mvSuSiE, flashfm, fastPAINTOR) under scenarios (b). For each combination of method and heritability, we present the mean value and the standard error from 100 repetitions. Panel A gives summaries of the power of detecting causal SNPs in at least one credible set. Panel B evaluates the coverage of credible sets, with the black dashed line corresponding to the 95% level. Panel C evaluates the average size of credible sets.
Fig 3.
Comparison of SuSiE and SuSiE2 with a smaller gene expression dataset and 1KG panel.
We simulated gene expression levels based on 2,000 UKBB samples, and simulated trait of interest based on 10,000 different UKBB samples under scenario (b). We compare the 95% credible sets from SuSiE and SuSiE2 from 100 repetitions. Panel A evaluates the power of detecting causal SNPs in at least one credible set. Panel B evaluates the coverage of credible sets, with the black dashed line corresponding to the 95% level.
Fig 4.
Average size of credible sets by SuSiE and SuSiE2.
We applied the SuSiE2 pipeline with gene expression from subcutaneous adipose (left) and visceral adipose (right). The 95% credible sets were grouped according to the chromosome in which they are located, labeled by chri for the ith chromosome. The minimum absolute correlation allowed in a credible set was fixed at 0.9 for both SuSiE and SuSiE2.
Table 1.
Summary of credible sets detected by SuSiE and SuSiE2 in the BMI study.
Fig 5.
Fine-mapping examples on BMI risk loci with SuSiE and SuSiE2.
The plots show the estimated PIPs for each SNP in two risk regions by SuSiE and SuSiE2. Panel A presents the results for a risk region on chromosome 1. Panel B illustrates the results for a risk region on chromosome 6. The SNPs from the same 95% credible sets by SuSiE or SuSiE2 are surrounded by circles in the same corresponding color. We label each credible set with the SNP ID of the leading variant.
Table 2.
Summary of AD mediators detected by SuSiE and SuSiE2.
Table 3.
Summary of credible sets identified by SuSiE and SuSiE2.
Fig 6.
Estimated PIP for each AD mediator by SuSiE and SuSiE2.
There were in total 20 AD risk loci divided into the following three categories. Five mediators were captured by both SuSiE and SuSiE2 in one credible set, denoted by the blue dots. SuSiE2 identified four additional risk loci, denoted by the green dots. The remaining 11 loci could not be captured in any credible set by either SuSiE or SuSiE2, corresponding to the red dots.
Fig 7.
Estimated PIPs by SuSiE, SuSiE2 and eQTL-based SuSiE for PICALM (A) and C14orf93 (B).
The PIPs estimated from the eQTL study are used as the prior information by SuSiE for SuSiE2. For the PICALM locus, PIPs for the true mediator in this locus are surrounded by the purple circle, and the points surrounded by an orange triangle correspond to the credible set from SuSiE2 which can capture the true mediator. For the C14orf93 locus, the true mediator was not included in the common part of summary statistics and ROSMAP data.