backwash and limma::limmaFit() gives identical outputs?

[jakeyeung]

Hi David,

Really cool R package, I am enjoying digging into it.

I noticed that backwash() gives identical output to limma::limmaFit() in the simulation example you have?

Is this expected behavior? If so, is there something in the backwash output that lets me know that it is not doing any hidden confounder adjustment?

I paste code example below.

Kindest regards,

Jake

rm(list=ls())

library(vicar)
library(ggplot2)
library(dplyr)
data(sim_gtex)
Y <- sim_gtex$Y
X <- sim_gtex$X
ctl <- sim_gtex$ctl
which_null <- sim_gtex$which_null
beta <- sim_gtex$beta

num_sv <- sva::num.sv(dat = t(Y), mod = X, method = "be")
bout.sim <- backwash(Y = Y, X = X, k = num_sv, cov_of_interest = 2, include_intercept = FALSE)

lmout.sim <- limma::lmFit(t(Y), X)
betahat_ols <- lmout.sim$coefficients[, 2]
sebetahat_ols <- lmout.sim$stdev.unscaled[, 2] * lmout.sim$sigma

# identical 
plot(bout.sim$result$betahat, betahat_ols)


head(bout.sim$result$betahat)
head(betahat_ols)

Kindest regards,

Jake

[dcgerard]

Hey @jakeyeung,

Hoo boy, this was awhile ago. But Backwash takes the OLS estimates as input to a model. So bout.sim$result$betahat, are the OLS estimates, which are the inputs, not the outputs. The posterior mean from that model's output is bout.sim$result$PosteriorMean. You can compare those to betahat_ols.

Also, Backwash does confounder adjustment. It assumes there are k confounders.

Cheers,
David

[jakeyeung]

Hi David,

Thanks so much for the clarification!

Is the OLS estimates in Backwash the same as Mouthwash? I noticed that the bout.sim$result$betahat would be slightly different than mout.sim$result$betahat (i.e. from mouthwash(Y = Y, X = X, k = num_sv, cov_of_interest = 2, include_intercept = FALSE))

Jake