Hi All,
I'm fitting brownian motion and OU models to gene expression data
(from 4 closely related species) and wondering if I need to mean
center (i.e. mean = 0) and scale (to variance = 1) the fitted gene
expression values prior to fitting the trait models.
I should add that our raw RNA-seq counts were normalized before
fitting a generalized linear model to obtain fitted expression
values. It is these fitted values that I used for BM and OU
estimation, after log2-transforming them.
I used Brownie to fit single mean OU models to all the genes in our
expression dataset (i.e. ~20K) and then followed the derivation in
Bedford and Hartl (PNAS 2009) to calculate selective constraints for
each gene. For the BM analyses, I used brownie.lite() (part of
phytools) for hypothesis testing of 1- and 2-rate BM models on the
differentially-expressed genes.
In Bedford and Hartl they mean centered and scaled each microarray's
raw microarray probe intensities as their normalization procedure.
Since we normalized the raw RNA-seq counts *before* fitting the linear
model and getting the fitted expression values, does it make sense to
mean center and scale our fitted values before fitting OU or BM models??
Sorry if this is obvious to you, but I'm not sure of what is the best/
proper way to do this. It is clear, though, that mean centering and
variance scaling would decrease expression divergence.
Any feedback would be greatly appreciated!!
Thanks,
Dan.
--------------------------------------------------------
Daniel Fulop, Ph.D.
Postdoctoral Scholar
Dept. Plant Biology, UC Davis
Maloof Lab, Rm. 2111
Life Sciences Addition, One Shields Ave.
Davis, CA 95616
530-752-8086
dfu...@ucdavis.edu
_______________________________________________
R-sig-phylo mailing list
R-sig-phylo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-phylo
