How to perform logistic regression (with option “firth-fallback”) in Plink2

Question

[MAPK2]

I am trying to perform this analysis as in this paper: https://doi.org/10.1002/alz.12319

"In the case-control replication data sets we performed a logistic regression (with option “firth-fallback”) for case/control status as implemented in PLINK 2. We included sex, age, sequencing center, and 5 Jaccard principal components with standardized variance as covariates."

So this is the command I used, but based on my QQ plot, this command doesn't look right. Can someone please suggest me if I am missing anything? @chrchang523 what would be your suggestion?

plink2 --bfile ${BFILE} --covar Phenoscope.txt --covar-name SEX,AGE,PC1,PC2,PC3,PC4,PC5 --covar-variance-standardize --pheno Phenoscope.txt --pheno-name STATUS --glm firth-fallback --allow-no-sex --out OUTPUT_${BFILE}

My QQ plot

LOG

PLINK v2.00a2.3LM AVX2 Intel (24 Jan 2020)
Options in effect:
  --allow-no-sex
  --bfile AQUILLA_Brian_2445_WXS_SNPS_INDELS_picard_biallelic-hwe-geno0.05-mind0.1-WXSm-SCOPEm_with_STATUS-clean3-geno-0.02-maxmaf-0.01
  --covar Phenoscope.txt
  --covar-name SEX,AGE,PC1,PC2,PC3
  --covar-variance-standardize
  --glm firth-fallback
  --out Manhattan_AQUILLA_Brian_2445_WXS_SNPS_INDELS_picard_biallelic-hwe-geno0.05-mind0.1-WXSm-SCOPEm_with_STATUS-clean3-geno-0.02-maxmaf-0.01
  --pheno Phenoscope.txt
  --pheno-name STATUS

Hostname: wustl.edu
Working directory: /06-Aquilla_202101-b/03-plink-QC-files
Start time: Mon Jul 19 01:03:16 2021

Note: --allow-no-sex no longer has any effect.  (Missing-sex samples are
automatically excluded from association analysis when sex is a covariate, and
treated normally otherwise.)
Random number seed: 1626674596
773748 MiB RAM detected; reserving 386874 MiB for main workspace.
Using up to 56 threads (change this with --threads).
1590 samples (762 females, 822 males, 6 ambiguous; 1590 founders) loaded from
AQUILLA_Brian_2445_WXS_SNPS_INDELS_picard_biallelic-hwe-geno0.05-mind0.1-WXSm-SCOPEm_with_STATUS-clean3-geno-0.02-maxmaf-0.01.fam.
1315729 variants loaded from
AQUILLA_Brian_2445_WXS_SNPS_INDELS_picard_biallelic-hwe-geno0.05-mind0.1-WXSm-SCOPEm_with_STATUS-clean3-geno-0.02-maxmaf-0.01.bim.
1 binary phenotype loaded (667 cases, 651 controls).
5 covariates loaded from Phenoscope.txt.
--covar-variance-standardize: 5 covariates transformed.
Calculating allele frequencies... done.
Warning: Skipping chrX in --glm regression on phenotype 'STATUS', since
correlation between covariates 'SEX' and 'SEX' is too high (CORR_TOO_HIGH).
--glm logistic-Firth hybrid regression on phenotype 'STATUS': done.
Results written to Manhattan_AQUILLA_Brian_2445_WXS_SNPS_INDELS_picard_biallelic-hwe-geno0.05-mind0.1-WXSm-SCOPEm_with_STATUS-clean3-geno-0.02-maxmaf-0.01.STATUS.glm.logistic.hybrid .

End time: Mon Jul 19 01:04:09 2021

[chrchang523]

There’s nothing obviously wrong in the command line. Can you post the full .log file from that run?

[MAPK2]

@chrchang523 Please see the updated post with full log file.

[chrchang523]

1. I don’t see PC4/PC5 in that log… and perhaps more importantly, a sequencing-center covariate.
2. To get results on chrX, use the —glm ‘sex’ modifier instead of including sex from the covariate file. (The problem is that, by default, .fam file sex is added as a covariate on chrX, and this conflicts with your manually added sex covariate.)
3. It probably won’t make a difference, but I’d use a newer plink2 build.