search

gusevlab / fusion_twas

methods for functional summary-based imputation
http://gusevlab.org/projects/fusion/
GNU General Public License v3.0
75 stars 45 forks source link

Uncertainty estimate? #26

Closed carbocation closed 4 years ago

carbocation commented 4 years ago

Running FUSION.assoc_test.R yields a Z score and a P value, but aside from the directionality yielded by the Z score, I interpret these as effectively yielding the same information. Is it possible to obtain either a standard error or a beta in the output (either of which, with a Z score, would permit calculation of the other)?

sashagusev commented 4 years ago

Hi, sorry about the delay in responding. All of the FUSION analyses are done on the variance/Z-score scale so the Z-score is the primary statistic being inferred since there is no underlying "allele" that's being tested. However, you could convert the Z-score back into a standardized effect-size by assuming the standard error of the underlying association is approximately \se = 1/sqrt(N) where N is the GWAS sample size (assuming this is in the tens of thousands and so the approximation is fairly robust), and then back-calculating \beta = Z * \se = Z/sqrt(N). This is an estimate of the \beta you would get from a regression of the phenotype on the centered and scaled predicted gene expression, if you were running the analysis with individual level data. Hope that helps!