Dataset for a Multivariate Genome-Wide Association Study of Psycho-Cardiometabolic Multimorbidity

Data collection method:

This dataset was created using genomic structural equation modeling (Genomic SEM) package in R. First, a common factor model was specified to capture the shared variance between major depression, coronary artery disease, and type 2 diabetes. The shared variance reflected the latent psycho-cardiometabolic multimorbidity factor. Subsequently, summary statistics for psycho-cardiometabolic multimorbidity were generated by regressing the latent factor on each SNP, resulting in 6,820,149 SNPs. Full details of the methodology may be found in the Methods section of the associated paper.

Technical details and requirements:

When using multimorbidity-associated variants as genetic instruments in Mendelian randomization analysis we recommend to only use non-heterogeneous SNPs. At the minimum, we recommend removing variants with QSNP P < 5e-8 and directionally discordant univariate effect estimates (option 1 below), as done in the non-heterogenous summary statistics versions (PCM_multimorbidity_summary_stats_noUKBB_non-het.txt.gz and PCM_multimorbidity_summary_stats_withUKBB_non-het.txt.gz). To ensure your analysis captures multimorbidity between coronary artery disease, type 2 diabetes and major depression, you may also apply a more stringent threshold for removing heterogeneous variants (e.g., QSNP P < 5e-6; option 2 below). A sample code for implementing this in R is included below: ``` # load tidyverse package library(tidyverse) # read in summary statistics sumstats <- read.delim("PCM_multimorbidity_summary_stats_withUKBB_non-het.txt.gz") # Option 1: remove discordant variants with QSNP P < 5e-8 data_not_concordant <- sumstats %>% filter(concordant == "FALSE") mm_snp_remove <- sumstats %>% filter(SNP %in% data_not_concordant$SNP) %>% filter(chisq_pval < 0.00000005) sumstas_nonhet <- sumstas %>% filter(!SNP %in% mm_snp_remove$SNP) # save file write.table(sumstas_nonhet, file = “newfile.txt", quote =F, sep = "\t", row.names = F) # Option 2: remove variants with QSNP P < 5e-6 sumstas_nonhet <- sumstas %>% filter(chisq_pval < 0.000005) # save file write.table(sumstas_nonhet, file = “newfile.txt", quote =F, sep = "\t", row.names = F) ```

Additional information:

For a detailed description of both sets of statistics, please see the associated publication. Summary statistics include: - SNP = single nucleotide polymorphism (SNP); - CHR = chromosome; - BP = position; - MAF = minor allele frequency; - A1 = effect allele; - A2 = non-effect allele; - BETA = effect of the SNP on the common factor; - SE = standard error; - Z_Estimate = ratio of effect/SE; - P = p-value of Z_Estimate; - chisq = chi-square statistic, providing the heterogeneity estimate for that SNP (equivalent to the Q SNP index); - chisq_df = chi-square degrees of freedom; - chisq_pval = chi-square p-value; - concordant = indicates whether effect estimates from three input genome-wide association studies were directionally concordant (TRUE = all three estimates were concordant; FALSE = at least one estimate was not concordant)

Methodology link: