h2 result with low mean(chi^2), low lambda GC (maybe), and high ratio
Yu FANG
Hi LDSC team,
I have been using ldsc to calculate h2 for quantitative trait of subjects with European ancestry (n~3500). In the result, the mean chi^2 is a little less than 1.02, lambda GC = 1.009. If I don’t constrain the intercept, the ratio is very high (0.74). After constraining intercept to 1, h2 was increased and SE was reduced - 0.16(0.12), but didn’t reach significance yet. I used only European ancestry samples and used top 10 PC as covariates in GWAS. Other than that I didn’t do extra GC corrections.
So my question is, is Lambda GC = 1.009 relatively small? Would it be possible that the high ratio is related to small lambda GC? I know the formula is (intercept-1)/(meanchi^2-1), obviously my mean chi^2 is small, but is that sort of related to small lambda GC?
On the other hand, I read from some post that if the data is GC corrected, the intercept might be negative, but it is not the case in my data. So I am wondering if my data is suitable for constraining the intercept?
I attach the log files of munge_sumstats and h2 calculation in the end. Could you help to interpret these results and give me some suggestions to improve the result (if possible)?
Thanks in advance for your help!
-Yu
Below are the log files:
1) the log file of munge_sumstats:
*********************************************************************
* LD Score Regression (LDSC)
* Version 1.0.0
* (C) 2014-2015 Brendan Bulik-Sullivan and Hilary Finucane
* Broad Institute of MIT and Harvard / MIT Department of Mathematics
* GNU General Public License v3
*********************************************************************
Call:
./munge_sumstats.py \
--out mydata\
--merge-alleles bp_w_hm3.snplist \
--N 3388.0 \
--sumstats mydata_gwas.txt
Interpreting column names as follows:
info: INFO score (imputation quality; higher --> better imputation)
snpid: Variant ID (e.g., rs number)
a1: Allele 1, interpreted as ref allele for signed sumstat.
pval: p-Value
beta: [linear/logistic] regression coefficient (0 --> no effect; above 0 --> A1 is trait/risk increasing)
a2: Allele 2, interpreted as non-ref allele for signed sumstat.
maf: Allele frequency
Reading list of SNPs for allele merge from bp_w_hm3.snplist
Read 1217077 SNPs for allele merge.
Reading sumstats from mydata_gwas.txt into memory 5000000 SNPs at a time.
Read 12185585 SNPs from --sumstats file.
Removed 10992633 SNPs not in --merge-alleles.
Removed 14 SNPs with missing values.
Removed 742716 SNPs with INFO <= 0.9.
Removed 3699 SNPs with MAF <= 0.01.
Removed 0 SNPs with out-of-bounds p-values.
Removed 149 variants that were not SNPs or were strand-ambiguous.
446374 SNPs remain.
Removed 0 SNPs with duplicated rs numbers (446374 SNPs remain).
Using N = 3388.0
Median value of beta was 0.000146605, which seems sensible.
Removed 46 SNPs whose alleles did not match --merge-alleles (446328 SNPs remain).
Writing summary statistics for 1217077 SNPs (446328 with nonmissing beta) to mydata_hm3.sumstats.gz.
Metadata:
Mean chi^2 = 1.01
WARNING: mean chi^2 may be too small.
Lambda GC = 1.009
Max chi^2 = 30.731
1 Genome-wide significant SNPs (some may have been removed by filtering).
Conversion finished at Tue Aug 8 17:25:55 2017
Total time elapsed: 2.0m:17.78s
2) the log file of h2 calculation without intercept constraint:
*********************************************************************
* LD Score Regression (LDSC)
* Version 1.0.0
* (C) 2014-2015 Brendan Bulik-Sullivan and Hilary Finucane
* Broad Institute of MIT and Harvard / MIT Department of Mathematics
* GNU General Public License v3
*********************************************************************
Call:
./ldsc.py \
--h2 mydata_hm3.sumstats.gz \
--ref-ld-chr bp_eur_w_ld_chr/ \
--out mydata_hm3_h2 \
--w-ld-chr bp_eur_w_ld_chr/
Beginning analysis at Tue Aug 8 17:26:30 2017
Reading summary statistics from mydata_hm3.sumstats.gz ...
Read summary statistics for 446328 SNPs.
Reading reference panel LD Score from bp_eur_w_ld_chr/[1-22] ...
Read reference panel LD Scores for 1293150 SNPs.
Removing partitioned LD Scores with zero variance.
Reading regression weight LD Score from bp_eur_w_ld_chr/[1-22] ...
Read regression weight LD Scores for 1293150 SNPs.
After merging with reference panel LD, 445427 SNPs remain.
After merging with regression SNP LD, 445427 SNPs remain.
Using two-step estimator with cutoff at 30.
Total Observed scale h2: 0.0413 (0.1864)
Lambda GC: 1.0075
Mean Chi^2: 1.0104
Intercept: 1.0077 (0.0099)
Ratio: 0.7406 (0.9448)
Analysis finished at Tue Aug 8 17:26:46 2017
Total time elapsed: 15.95s
3) the log file of h2 calculation with constraining intercept to 1:
*********************************************************************
* LD Score Regression (LDSC)
* Version 1.0.0
* (C) 2014-2015 Brendan Bulik-Sullivan and Hilary Finucane
* Broad Institute of MIT and Harvard / MIT Department of Mathematics
* GNU General Public License v3
*********************************************************************
Call:
./ldsc.py \
--h2 mydata _hm3.sumstats.gz \
--ref-ld-chr bp_eur_w_ld_chr/ \
--intercept-h2 1 \
--out mydata _hm3_int1_h2 \
--w-ld-chr bp_eur_w_ld_chr/
Beginning analysis at Tue Aug 8 17:27:27 2017
Reading summary statistics from mydata _hm3.sumstats.gz ...
Read summary statistics for 446328 SNPs.
Reading reference panel LD Score from bp_eur_w_ld_chr/[1-22] ...
Read reference panel LD Scores for 1293150 SNPs.
Removing partitioned LD Scores with zero variance.
Reading regression weight LD Score from bp_eur_w_ld_chr/[1-22] ...
Read regression weight LD Scores for 1293150 SNPs.
After merging with reference panel LD, 445427 SNPs remain.
After merging with regression SNP LD, 445427 SNPs remain.
Total Observed scale h2: 0.1601 (0.1178)
Lambda GC: 1.0075
Mean Chi^2: 1.0104
Intercept: constrained to 1.
Analysis finished at Tue Aug 8 17:27:40 2017
Total time elapsed: 12.99s