[R-meta] rma.mv metafor models for genome-wide meta-analyses are (too) conservative

[St Pourcain, Beate]

Hi,

We are a group of geneticists using meta-regression for genome-wide
meta-analysis and encountered a hidden thorny issue.

We use metafor rma.mv models to meta-analyse:

 

*	130 correlated input statistics 
*	each statistics has BETAs for 8 million variants (reflecting genetic
association effects) 

 

capturing:

*	70000 individuals
*	400000 repeat observations

 

from

23 cohorts.

 

Across the 8 million variants, the following model fits quite well for each
variant, based on a (fairly) well-known phenotypic correlation matrix for
sample overlap (Vsampoverlap) scaled to the SE of each BETA.

 

model_cov <- rma.mv(yi = BETA, V = Vsampoverlap, mods = COV1+ COV2+COV3,
random= list(~ 1|COHORT/VAR), data = df)}, silent = F)

 

where VAR specifies the input statistics, COHORT represents cohorts, and COV
represents fixed effects.

 

However, when predicting BETAs for each of the 8 million variants across a
grid of fixed effect predictors (COV1, COV2, COV3) we note that, on average,
derived genome-wide Z-scores based on predicted BETAs and SEs are too
conservative and deviate from the expected null distribution in a
quantile-quantile plot, affecting all predictions. We can also quantify this
deviation from a null distribution across all 8 million variants using the
LDSC intercept (i.e. the Linkage disequilibrium Score regression intercept
for a variant-based heritability estimation; Bullik-Sullivan 2015:
https://pubmed.ncbi.nlm.nih.gov/25642630/), which should be one if unbiased,
but we observe ~0.9(SE=0.005). Note that the intercept would be above one,
if test statistics were inflated.

 

Thus, we think we subtly over-correct for relatedness and quench the power
of our analysis. Would there be any thoughts within the group as to how to
relax the adjustment for relatedness in metafor?

 

Thanks so much for any input,

Beate

 

PS: Note to random geneticists in the group: If we could, we would replace
the phenotypic correlation matrix with the non-genetic part of the
phenotypic correlation matrix, but such an estimation is unreliable given
the small sample size of most cohorts.

 

 

R version 4.1.3 (2022-03-10)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19045)
 
Matrix products: default
 
locale:
[1] LC_COLLATE=English_Europe.1252  LC_CTYPE=English_Europe.1252
LC_MONETARY=English_Europe.1252 LC_NUMERIC=C                   
[5] LC_TIME=English_Europe.1252    
 
attached base packages:
[1] parallel  stats     graphics  grDevices utils     datasets  methods
base     
 
other attached packages:
[1] data.table_1.14.2   metafor_4.4-0       numDeriv_2016.8-1.1
metadat_1.2-0       Matrix_1.4-0 

 

 

Beate St Pourcain, PhD

Senior Investigator & Group Leader

Room A207

Max Planck Institute for Psycholinguistics | Wundtlaan 1 | 6525 XD Nijmegen
| The Netherlands

 

 

@bstpourcain

Tel:  <tel:+31%2024%20352%201964> +31 24 3521964

Fax:  <tel:+31%2024%20352%201213> +31 24 3521213

ORCID:  <https://orcid.org/0000-0002-4680-3517>
https://orcid.org/0000-0002-4680-3517

Web:
<https://www.mpi.nl/departments/language-and-genetics/projects/population-va
riation-and-human-communication/>
https://www.mpi.nl/departments/language-and-genetics/projects/population-var
iation-and-human-communication/

Further affiliations with:

MRC Integrative Epidemiology Unit | University of Bristol | UK

Donders Institute for Brain, Cognition and Behaviour | Radboud University |
The Netherlands

 

My working hours may not be your working hours. Please do not feel obligated
to reply outside of your normal working schedule.

 


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20240126/db7ad2ed/attachment-0001.html>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 7322 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-meta-analysis/attachments/20240126/db7ad2ed/attachment-0001.p7s>