For discovery gene-based meta-analyses, we utilised three statistical methods as part of the RAREMETAL package: the Weighted Sum Test (WST) [26], the burden test [27] and the Sequence Kernel Association test (SKAT) [28]. EPACTS (v.3.3.0) [29] was used to annotate variants (for use in gene-based meta-analyses), as recommended by RAREMETAL. Two MAF cut-offs were used, one used low-frequency (MAF < 0.05) and rare variants, the second only used rare variants (MAF < 0.01). Nonsynonymous, stop gain, splice site, start gain, start loss, stop loss, and synonymous variants were selected for inclusion. A sensitivity analysis to exclusion of synonymous variants was also performed. Gene-level associations with P < 8 × 10−7 were deemed statistically significant (Bonferroni-adjusted for ~20,000 genes and three tests at α = 0.05). To examine if the gene associations were driven by a single variant, the gene tests were conducted conditional on the SNV with the smallest P-value in the gene, using the shared single variant association statistic and covariance matrices [21, 25].
