Open MarioGuCBMR opened 4 years ago
What happens if you calculate the z-score for your summary statistics and use the Z/N specification?
On Mon, May 25, 2020 at 1:15 PM MarioGuCBMR notifications@github.com wrote:
Hi, I am comparing two big GWAS from the same paper. They are both big meta-analysis with the same sample, but from two different traits. For both sets of SNPs I have 27.000.000 SNPs so I run out of memory. Hence, I decided to take a small portion of the SNPs: 180.000, since it was the proportion of SNPs used in the original paper. I took this SNPs randomly. In theory, this should work just fine, right? I am only testing if MTAG does work.
My results end up worsening some p-values. I had 120 Genome-Wide significant SNPs and I end up getting around 60 for the first trait. I went to check the second, and the amount of Genome-Wide significant SNPs improved a lot. I think that is to be expected, due to the nature of the traits, though. However, in the results I do get strange figures that make me think that MTAG might have not run correctly:
Trait # SNPs used N (max) N (mean) GWAS mean chi^2 MTAG mean chi^2 GWAS equiv. (max) N 1 MTAG1.txt 153638 697693 498364 1.014 0.897 -5023474 2 MTAG2.txt 153638 806826 505661 1.782 1.599 618287
The first things was the GWAS equivalent max N, which ends up being negative!
Estimated Omega: [[-2.091e-07 5.282e-11] [ 5.282e-11 1.354e-06]]
(Correlation): [[nan nan] [nan 1.]]
And the second one is this correlation matrix.
Estimated Sigma: [[1.619 0.258] [0.258 1.024]]
(Correlation): [[1. 0.201] [0.201 1. ]]
MTAG weight factors: (average across SNPs) [0.842 0.835]
For both traits these are the headers that I used:
chr pos snpid a1 a2 freq beta se pval n 1 2957600 rs12409277 C T 0.1904 0.0136 0.0023 3.321e-09 630061 1 9329289 rs2071931 T C 0.2146 0.0165 0.0021 1.23e-14 630042 1 9346583 rs72642703 G A 0.8747 0.0183 0.0029 5.377e-10 485486
And these are the commands:
python mtag.py --sumstats MTAG2.txt,MTAG2.txt --out ./WHR_BMI_results --n_min 0.0 --use_beta_se --beta_name beta --se_name se --stream_stdout
I hope you know how to interpret these results! When I increase the number of SNPs (500.000) it raises an error when trying to calculate the standardized betas since it says that some SEs are 0, when they are not.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/omeed-maghzian/mtag/issues/95, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFBUB5IFHTH6YPUVGOYO5F3RTKRULANCNFSM4NJVN6CQ .
Hi, I am comparing two big GWAS from the same paper. They are both big meta-analysis with the same sample, but from two different traits. For both sets of SNPs I have 27.000.000 SNPs so I run out of memory. Hence, I decided to take a small portion of the SNPs: 180.000, since it was the proportion of SNPs used in the original paper. I took this SNPs randomly. In theory, this should work just fine, right? I am only testing if MTAG does work.
My results end up worsening some p-values. I had 120 Genome-Wide significant SNPs and I end up getting around 60 for the first trait. I went to check the second, and the amount of Genome-Wide significant SNPs improved a lot. I think that is to be expected, due to the nature of the traits, though. However, in the results I do get strange figures that make me think that MTAG might have not run correctly:
Trait # SNPs used N (max) N (mean) GWAS mean chi^2 MTAG mean chi^2 GWAS equiv. (max) N 1 MTAG1.txt 153638 697693 498364 1.014 0.897 -5023474
2 MTAG2.txt 153638 806826 505661 1.782 1.599 618287
The first things was the GWAS equivalent max N, which ends up being negative!
Estimated Omega: [[-2.091e-07 5.282e-11] [ 5.282e-11 1.354e-06]]
(Correlation): [[nan nan] [nan 1.]]
And the second one is this correlation matrix.
Estimated Sigma: [[1.619 0.258] [0.258 1.024]]
(Correlation): [[1. 0.201] [0.201 1. ]]
MTAG weight factors: (average across SNPs) [0.842 0.835]
For both traits these are the headers that I used:
chr pos snpid a1 a2 freq beta se pval n 1 2957600 rs12409277 C T 0.1904 0.0136 0.0023 3.321e-09 630061 1 9329289 rs2071931 T C 0.2146 0.0165 0.0021 1.23e-14 630042 1 9346583 rs72642703 G A 0.8747 0.0183 0.0029 5.377e-10 485486
And these are the commands:
python mtag.py --sumstats MTAG2.txt,MTAG2.txt --out ./WHR_BMI_results --n_min 0.0 --use_beta_se --beta_name beta --se_name se --stream_stdout
I hope you know how to interpret these results! When I increase the number of SNPs (500.000) it raises an error when trying to calculate the standardized betas since it says that some SEs are 0, when they are not.