NEWS.md
This version mostly changes the documentation. Additions include:
vruv4
, ruv3
, ruvb
, mouthwash
, and backwash
.ruvb
, and one giving a sample analysis using the functions in vicar and other confounder adjustment packages. Type vignette(package = "vicar")
for the list of available vignettes.fa_tester
and fa_tester_ruvb
for testing whether a user-specified factor analysis is appropriate for the functions in vicar.data(sim_gtex)
to access the data or ?sim_gtex
for seeing details about the data.vruv2
from being exported as it was supplanted by ruv3
.This version added the function backwash
. This is very similar in spirit to mouthwash
, except that rather than estimate the confounders by maximum likelihood, backwash
does so using a more Bayesian approach. backwash
returns a variational approximation to the posterior.
This version added the function mouthwash
to adjust for hidden confounding when one does not have control genes. It applies the same prior from ashr to a factor-augmented regression framework.
A lot of changes have occurred since my last news update. The biggest changes are:
ruv3
.ruvimpute
is a generic function for using matrix imputation for confounder adjustment.ruvb
is a special Bayesian version of RUV. It is highly customize-able, as you can tweak the Bayesian factor analysis and the prior specifications of the effects.vruv2
as this is now subsumed by ruv3
. I’ll probably remove vruv2
in the future.I provide reasonable defaults for all new methods.
vruv2
now works pretty well and is recommended for use. This is a very different way to do variance inflation in RUV2 than what was previously implemented.vruv2_old
, but it may be removed at any time.ash_ruv2
that is a wrapper for vruv2
and ashr::ash.workhorse
.pca_ruv2
, qmle_ruv2
, and pca_2step
. In the future, I plan on only saving pca_2step
.vruv2
, a variance-inflated version of RUV2, but it doesn’t work too well yet.vruv4
. I thought that vicarius_ruv4
was too verbose. In the future, as I create new calibrated versions of confounder adjustment methods, the function name will just have a “v” in front of the name of the confounder adjustment method.vruv4
now multiplies the estimated variances by ([X, Z]’[X, Z])^{-1} rather than (X’X)^{-1}.vruv4
when k = 0
. Keep this in mind.limmashrink = TRUE
is now the default.