cellHTS2 - normalization with BScore (performance
1
0
Entering edit mode
liebi83 • 0
@f3284e1c
Last seen 2.3 years ago
Germany

Dear Support, I am having "issues" with regards of performance of the normalizePlates function

My current setup: I have 100 plates (384 dimension; replicate = 1; channel = 1) of a screening campaign and I would like to use BScore as we are facing side effects on the plates. I am using follwoing command to normalize the data

xn <- normalizePlates(x, scale="multiplicative", log=FALSE, method="Bscore", varianceAdjust="byPlate", save.model=TRUE)

The function call takes very long. In my case with 100 plates: ~ 40 minutes. Can this be done quicker? Thanks for any hints! Alex

cellHTS2 Normalization • 886 views
ADD COMMENT
0
Entering edit mode
@wolfgang-huber-3550
Last seen 3 months ago
EMBL European Molecular Biology Laborat…

Fitting a B-Score model should certainly not take 24 seconds per plate, the actual maths is probably less than 1% of that, with the rest going to data shuffling and copying. The package is rather old and convoluted.

I think the modern way of doing things is simply to put the screen data into a big data.frame or tidytable and use dplyr, group_by etc. for such operations. It should be simple to implement B-score this way, maybe someone has already done it?

You could also have a look at Junyan Lu's DrugScreenExplorer package (https://lujunyan1118.github.io/DrugScreenExplorer ) and in particular the fitEdgeEffect function https://github.com/lujunyan1118/DrugScreenExplorer/blob/master/R/processData.R Also note that we have found fitting a smooth loess or sigmoid surface better than the B-score for dealing with spatially correlated effects: they use fewer free parameters (degrees of freedom) than the B-score and thus are less prone to overfitting (the potential flipside of doing edge correction).

ADD COMMENT

Login before adding your answer.

Traffic: 713 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6