We conducted TWAS/PWAS/spliceWAS using the FUSION package (Gusev et al., 2016). Specifically, the FUSION.assoc_test.R script was utilised, with the best performing GReX model selected from five-fold cross-validation (R2) and the SNP weight set selected by FUSION for calculating the TWAS Z score. In accordance with usual practice for the FUSION approach, only genes/transcripts with significantly non-zero cis-acting heritability (cis-h2) are included. Prior to analysis, summary statistics were also munged whereby SNPs were retained with an imputation INFO > 0.9, as well as removing indels, strand ambiguous SNPs and SNPs with MAF < 0.01. We corrected for the number of non-missing models tested for the TWAS, PWAS, and spliceWAS independently. Our primary method for correcting for multiple testing was the Bonferroni approach considering all models tested for each modality (TWAS, PWAS, and spliceWAS independently), however, this is inherently conservative due to genes having a GReX model in multiple tissues and correlations between genes. As a result, we also used a more exploratory Benjamini-Hochberg false discovery rate (FDR) approach for the TWAS, PWAS, and spliceWAS separately.
