zhanxw / rvtests

Rare variant test software for next generation sequencing data
131 stars 41 forks source link

Missing col error in version: 20170228 #29

Closed SailajaVeda closed 7 years ago

SailajaVeda commented 7 years ago

I am using RVtest Program version: 20170228 and creating a covariance matrix with window size of 500KB, to reduce the size of the matrix. I see the following error in my log files, but the analysis runs to completion.

[ERROR] Missing 'u1141' column! [WARN] Failed to load spectral decomposition results of the kinship matrix! [INFO] DONE: Spectral decomposition of the kinship matrix succeeded in [ 3.4 ] seconds.

Wondering if anyone else has seen this and if this could be a reason why my Covariance matrices are > 400GB (for HRC imputed data) in spite of the requested 500kb window size. Thanks

zhanxw commented 7 years ago

This error happens when the decomposed kinship file does not match the input VCF file. Did RVTESTS stop after this error popped up?

SailajaVeda commented 7 years ago

I am sure that I have used the same vcf file to make the kinship matrix file but using a MAF filter. Yes the RVTESTS does complete the analysis after the error. The covariance matrix files are really large, ~450GB for HRC imputed data set in 3K samples.

dajiangliu commented 7 years ago

the covariance matrix size depends directly on the number of markers in a window. Sample sizes are not directly affecting the size of the cov file. Yet, typically, larger datasets come with larger number of segregating markers. Just FYI.

SailajaVeda commented 7 years ago

The study has ~15 m polymorphs. As the analyses are running to completion, is it safe to ignore the error?

zhanxw commented 7 years ago

I think it's safe to ignore the error. This error happens when the pre-calculated decomposed kinship cannot be used in the analysis. If that happens, RVTESTS just try to decompose the kinship and the computation result should be fine.

zhanxw commented 7 years ago

In the new version, I will make this error message more explicit. Thanks.