Recessive GWAS results format

R12 recessive results file format

This page was last updated for R12.

General overview

We provide recessive GWA results for 2,469 of release 12's (R12) endpoints. The analysis was performed with REGENIE v3.3. For variants with alternate allele frequency larger than 0.5, the effect allele was switched to be the reference allele, in order to keep the effect allele's frequency below 0.5. For endpoints that included female samples, only female samples were used in chromosome X analysis. After the REGENIE analysis, recessive results were joined with earlier additive model results.

Data locations

Recessive REGENIE files

Tabix-indexed REGENIE results are available for R12 at the bucket location: gs://finngen-production-library-green/finngen_R12/finngen_R12_analysis_data/summary_stats_recessive/results

with 2 files per endpoint (PHENO):

Column	Description
#chrom	Chromosome
pos	Chromosomal position (GRCh38)
ref	Reference allele
alt	Alternative allele
pval_add	Additive association p-value
mlogp_add	Additive association -log10(p-value)
beta_add	Per-allele additive model effect size estimate on endpoint for alt allele
sebeta_add	Standard error of additive model effect size estimate
af_alt	Allele frequency of alt allele
af_alt_cases	Allele frequenct of alt allele in cases only
af_alt_controls	Allele frequenct of alt allele in controls
pval	Recessive association p-value
mlogp	Recessive association -log10(p-value)
beta	Effect size of recessive genotype for effect allele
sebeta	Standard error of recessive model effect size estimate
effect_allele	Effect allele used in recessive association analysis
flipped	If the allele frequency was >0.5, the reference allele was used as the effect allele. In that case, this value is 'True', otherwise 'False'
mlogp_diff	difference of -log10(p-value) between additive and recessive models
INFO	Variant INFO score
AC_Het	Heterozygote allele count
AC_Hom	Homozygote allele count
most_severe	Most severe consequence of the alternate allele
gene_most_severe	Gene in which most severe consequence is in
rsid	variant rsid(s)
b37_coord	variant coordinate (chrom:pos:ref:alt) in build GRCh19
EXOME_enrichment_nfsee	variant enrichment in Finnish population vs non-Finnish European (NFE) population. Based on gnomAD 2.1 genome data.
EXOME_enrichment_nfe	variant enrichment in Finnish population vs non-Finnish European (NFE) population. Based on gnomAD 2.1 genome data.
GENOME_enrichment_nfee	variant enrichment in Finnish population vs non-Finnish European (NFE) population. Based on gnomAD 2.1 genome data.
GENOME_enrichment_nfe	variant enrichment in Finnish population vs non-Finnish European (NFE) population. Based on gnomAD 2.1 genome data.
index	Index of variant in variant annotation file

Methods overview

Genotype data split into AF<0.5 and AF>0.5 datasets

We identified the variants with alternate alelle frequency larger than 0.5, and separated those variants to its own dataset.

Recessive REGENIE GWAS analysis

We performed standard GWA analyses using REGENIE (Mbatchou, J. et al., 2021) v3.3 for each endpoint separately. In the model-building step (step1), the standard set of FinnGen covariates (age, 10 genetic principal components, genotyping chip and legacy batch number) were used. The exceptions were endpoints E4_PORPHYNAS and Q17_MARFAN where legacy batch number was not included as a covariate.

A subset of the imputed variants with INFO score > 0.95 in all batches, <3% missingness and MAF > 1% were LD pruned with a 1.5Mb window and an r2 threshold of 0.2 to leave 188,153 well-imputed independent variants to build the model. A block size of 1,000 variants was used in step 1.

For all 2,469 included endpoints, recessive model association testing (in step 2 of REGENIE) was performed on variants with a minimum allele count (MAC) of 5 in the set of samples with non-missing data. As with the standard FinnGen R12 REGENIE GWAS runs, the approximate Firth test was applied to variants that had an initial P<0.01, with the standard error estimated from the effect size and the likelihood ratio test p-value (i.e. using flags --firth --approx --pThresh 0.01 --firth-se). In step 2, we used a block size of 400. For the genotype dataset with alternate allele frequency greater than 0.5, we omitted the regenie option --ref-first, forcing regenie to consider the reference allele as the effect allele for those variants.

The REGENIE results files were then filtered, removing variants with -log10(P)="NA" (where the Firth approximation had failed) .