Closed Ojami closed 2 years ago
Hi @Ojami,
Thanks for all the information! When only QC'ed samples were used, are all 13 markers having MAF > 0. Do they have MAF > 1.176969e-05 (MAC > 0.5)? It seems only 6 of them passed the cutoff 1.176969e-05 <= MAF <= 0.01 and all of them have MAC <= 10.
When there is only 1 marker, only one p-value is output because it does not make sense to conduct different set-based tests. When lowering MACCutoff_to_CollapseUltraRare, not all ultra-rare markers are collapsed and there will be more than one marker, so you will see p-values for SKAT and Burden tests.
Thanks, Wei
Hi @weizhouUMICH
Thanks for your response. Please note that
21241 samples have been used to fit the glmm null model
and 21241 is a subset of both full (200643) and QCed (177690) samples, so both analyses should end up in the same results, simply because 21241 samples have been used for null model.
But SAIGE computes MAFs using VCF samples:
std::size_t sample_size = marker_file.samples().size();177690
Shouldn't MAFs correspond for samples used in step 1 (null)? because after all, these samples ( 21241 in my example) will be used for gene-/region-based tests.
Thanks! Oveis
Hi @Ojami,
Yes, You are right. The MAFs should always be calculated for samples used in step 1. Let me try to replicate the issue.
Thanks, Wei
Hi @Ojami,
Sorry for my very late reply! We have just released a new version 1.0.0. It has computational efficiency improvements for both Step 1 and Step 2 for single-variant and set-based tests. We have created a new program github page https://github.com/saigegit/SAIGE with the documentation provided https://saigegit.github.io/SAIGE-doc/ The program will be maintained by multiple SAIGE developers there. The docker image has been updated. Please feel free to try the version 1.0.0 and report issues if any.
Thanks! Wei
I was trying to perform SAIGE-GENE (
0.44.6.1
) step 2 with VCF files with the same samples used in constructing the GRM and null fitting. The results were not consistent with what I already knew from running SKAT (in R). Strictly speaking, in step 2 number of markers were lower than expected. So, I repeated the same analysis this time using a VCF file with all available samples (not QCed). This time the results were comparable. I post the log of these two scenarios (truncated) below.scenario 1: QCed VCF (same samples used for steps 0 and 1)
and results:
scenario 2: Full VCF (all samples)
and results:
Please note that the difference between number of markers in two scenarios is not due to 0 MAF, meaning those 13 markers in scenario 2 also have a non zero MAF in scenario 1. On a different note, SAIGE-GENE+ does not generate summary stats for SKAT and burden (all NaN). This is a direct consequence of
MACCutoff_to_CollapseUltraRare
cutoff value (lowering this value would force SAIGE to generate summary stats for SKAT and burden as well).Appreciate any thoughts in advance. OJ