METAL-Random

Question

[kris]

Hi, Can anyone please explain what the headers in the METAL-random (https://github.com/explodecomputer/random-metal) output mean? The header after i did random effect inverse variance meta-analysis in METAL-random gives the following:

MarkerName Allele1 Allele2 Freq1 FreqSE MinFreq MaxFreq Effect StdErr Pvalue Direction HetISq HetChiSq HetDf HetPVal EffectARE StdErrARE PvalueARE tausq StdErrMRE PvalueMRE

I am confused what PvalueARE and PvalueMRE mean and which is the p value to refer to for the random effect model ? I checked the METAL website and the github page but couldn't find much info regarding the random effect output. Pvalue i know should be for the fixed effect.

Thank you.

[Basti]

Information from here : https://github.com/MRCIEU/godmc_phase2_analysis

Fixed effects Effect StdErr pval

Heterogeneity stats : Direction HetISq HetChiSq (This is Q statistic) HetDf HetPVal tausq

Additive random effects (DerSimonian-Laird estimator) : EffectARE StdErrARE PvalueARE

Multiplicative random effects (The Effect size for this is the same as the Fixed effects) : StdErrMRE PvalueMRE

[kris]

Thanks for the link.

[Alessandro]

Hi, posting this question here since it is relevant to METAL-Random: I am trying to carry out a random effect meta-analysis, but for some reason get exactly the same results as fixed effect meta-analysis and don't get the additional (expected) columns mentioned above. I guess the usual inverse variance weighted Fixed Effect meta-analysis was run.

Posting the script below for clarity:

# === GENERIC SETTINGS of input files === #
GENOMICCONTROL ON
SEPARATOR TAB
MARKER cptid
ALLELE EFFECT_ALLELE OTHER_ALLELE
FREQ EAF
EFFECT BETA
PVALUE PVAL 
STDERR SE
WEIGHT N
STRAND STRAND
USESTRAND ON
AVERAGEFREQ ON
MINMAXFREQ ON
# === SETTINGS for the analysis === #
SCHEME STDERR
CUSTOMVARIABLE TotalSampleSize   # only when scheme stderr, otherwise Weight will be in the output file
LABEL TotalSampleSize as N
# === PROCESS THE  INPUT FILES === #
PROCESS a.txt.gz
PROCESS b.txt.gz
PROCESS c.txt.gz
PROCESS d.txt.gz
PROCESS e.txt.gz
PROCESS f.txt.gz
PROCESS g.txt.gz
PROCESS h.txt.gz
PROCESS i.txt.gz
PROCESS j.txt.gz
# === ANALYSIS === #
ANALYZE RANDOM

The only option I replaced was the last row (where ANALYZE HETEROGENEITY was standing before). Maybe some of my options are overriding the last command?

Hope you can help me.

Rgds, Alessandro

[Basti]

I am not aware of a ANALYZE RANDOM command in the METAL software. Random effect model is normally run with ANALYZE HETEROGENEITY. There may be a confusion with this modified version of METAL from GitHub : https://github.com/explodecomputer/random-metal

Pay attention that you installed/used the version you needed

[Alessandro]

Hi, actually that is the one I used, but no changes..

[Basti]

I think I would just go with ANALYZE HETEROGENEITY because it is the official version and maybe there are issues with the custom version found on GitHub, and I currently don't see the added value for random meta-analysis because it is already implemented in the official version

Answer 1

[Alessandro]

Hi, just a small update: I just downloaded the whole zipped folder from https://zenodo.org/record/6974696 and now it seems to work! Probably something went wrong in the first attempt. Thanks for the useful suggestions anyway!