From listmaster animalgenome.org Tlistmaster Feb 21 15:07:14 2019
Return-Path: <listmasteranimalgenome.org>
From: Andres Legarra-Albizu <andres.legarrainra.fr>
Subject: Re: software for exact p-values for association
analysis on single step gblup
Postmaster: submission approved by list moderator
To: Members of AnGenMap <angenmapanimalgenome.org>
Date: Thu, 21 Feb 2019 15:07:14 -0600
Hi Yuri and Dan
the p-values that we propose have nothing to do with “the variance
contributed by a marker”. We do use a studentized t-test. In a series
of papers (mentioned in our draft) with, among other people, Juan Pedro
Steibel and Rob Tempelman, they proved that the studentized t-test
can be obtained backsolving results from GBLUP *with identical results
to a typical GWAS by regression test*. These papers have been published
in the last years:
Chen, C., J. P. Steibel, and R. J. Tempelman, 2017 Genome-Wide Associ-
ation Analyses Based on Broadly Different Specifications for Prior Dis-
tributions, Genomic Windows, and Estimation Methods. Genetics 206:
1791–1806. https://doi.org/.../genetics.117.202259
Gualdrón Duarte, J. L., R. J. Cantet, R. O. Bates, C. W. Ernst, N. E.
Raney et al., 2014 Rapid screening for phenotype-genotype associations
by linear transformations of genomic evaluations. BMC Bioinformatics 15:
246. https://doi.org/10.1186/1471-2105-15-246
Bernal Rubio, Y. L., J. L. Gualdrón Duarte, R. O. Bates, C. W. Ernst, D.
Nonneman et al., 2016 Meta-analysis of genome-wide association from genomic
prediction models. Anim. Genet. 47: 36–48. https://doi. org/10.1111/age.12378
in these papers the theory was developed.
Then there is our own, who does a wrap-up of the theory and summarizes:
Legarra, A., Ricard, A., & Varona, L. (2018). GWAS by GBLUP: Single and
Multimarker EMMAX and Bayes Factors, with an Example in Detection of a
Major Gene for Horse Gait. G3: Genes, Genomes, Genetics, 8(7), 2301-2308.
I hope this helps
Andres
On 20 Feb 2019, at 17:21, DANIEL GIANOLA <gianolaansci.wisc.edu> wrote:
Yuri,
Your point is well taken, but the authors do recognize in the manuscript
that there are no theoretical thresholds for the distribution of a test
statistic resulting from dividing the effect estimate by the square root
of "2pq*alphahatsquared". More fundamentally, it seems to me that the
"the variance contributed by a marker" cannot be taken as an estimate
of the precision of estimation, contrary to the situation in standard
t-tests, as you correctly observe. Hence, their studentized criterion
lacks formal justification, it appears.
Perhaps the authors can provide some intuitively persuasive arguments.
Regards.
Daniel
________________________________
.From: Yuri Tani Utsunomiya <ytutsunomiyagmail.com>
.Sent: Wednesday, February 20, 2019 8:43 AM
.To: Members of AnGenMap <angenmapanimalgenome.org>
.Subject: Re: software for exact p-values for association analysis on single
step gblup
Dear Andres,
I am a little bit confused with the proposed test.
As far as my limited knowledge goes, classic t-distributed statistics are
ratios between estimates of parameters of interest and their respective
standard errors. In the manuscript, the denominator of the ratio seems to
be the square root of the variance due to the tested marker instead (i.e.,
marker standard deviation), which is not the same as the standard error of
the estimate. The latest information I had was that the standard errors of
marker effects in GBLUP and ssGBLUP do not have a simple or even known
expression for fast computation, and therefore they should be approximated
with resampling procedures such as bootstraping, subsampling or
jackknifing. Please do not take this observation as a critic, but rather as
a question that - I believe - many of us users of blupf90 have. Last but
not least, I apologize if my read of the proposal is incorrect.
Cheers,
Yuri Utsunomiya
On Wed, Feb 20, 2019 at 10:36 AM Andres Legarra-Albizu <andres.legarrainra.fr> wrote:
Dear all,
thanks to collaborative work and extensive programming, I, on behalf of all
the blupf90 team, am happy to announce that the blupf90 suite now computes
exact p-values for gwas analysis - same as “one at a time” classical
gwas by regression. The authoritative reference is this biorxiv deposit:
MS ID#: BIORXIV/2019/555243
MS TITLE: Exact p-values for large-scale single step genome-wide
association, with an application for birth weight in American Angus
These requires the use of OPTION snp_p_value in programs blupf90 (or
blupf90test) and postGSf90. More details in the corresponding sections in
the wiki:
http://nce.ads.uga.edu/...application_programs
regards, Andres
--
Andres Legarra
+33 561285182
INRA, UMR1388 GenPhySE
CS 52627
31326 Castanet Tolosan, France
http://genoweb.toulouse.inra.fr/~alegarra
|
