search

MarioniLab / PlateEffects2016

Code and manuscript files for Aaron's plate effect paper.
8 stars 3 forks source link

Mention correlation in real data #4

Closed LTLA closed 8 years ago

LTLA commented 8 years ago

We can put in the consensus correlation from duplicateCorrelation from the real data, to show that the plate effect actually exists and contributes to the total variance of the experiment.

LTLA commented 8 years ago

For the mESC data set, running:

cor <- duplicateCorrelation(v.all, design, block=paste0(targets$Batch,targets$Serum))

... yields a cor$consensus of ~0.27. Which is not huge, but it's definitely substantial, and it indicates that more than a quarter of the total variance in the experiment is attributable to the plate effect. This is comparable to the simulation setup in the paper; the variance in log-counts with a dispersion of 1 is around 1.6, and the variance of the log-plate effect is around 0.5.

LTLA commented 8 years ago

Rejigged the simulations based on an estimated variance, so we don't need this anymore.