josefin-werme / LAVA

51 stars 9 forks source link

Zero p-values in run.univ at non GWAS significant loci. #47

Closed gouthamatla closed 1 year ago

gouthamatla commented 1 year ago

Hi ,

Thanks for developing this excellent tool. I am looking at the run.univ() output. I noticed that I get too many loci to be significant (p<2e-5). When I inspected, most of those loci (with p==0) do not have any genome-wide or marginally significant SNPs for that trait in that locus. May I ask why this happens ? I can work on loci that has at least one GWAS sígnicant SNP but would like to know why this happens.

cadeleeuw commented 1 year ago

Hi,

Strong (and more or less global) inflations of the univariate signal from the LAVA results, relative to the input GWAS, generally are indicative of either a) something going wrong with the alignment of the SNP effects with the reference data, or b) a significant mismatch between the summary statistics and the reference data in terms of ancestry/LD.

My suggestion would be to start by verifying that the allele columns in the summary statistics provided to LAVA are correct, ie. that these consistently indicate the effect allele for that SNP (rather than eg. the minor allele). An input issue of this nature is usually the cause of this sort of problem.

Best, Christiaan