[BioC] Definition of the CellHTS2 normalizePlates parameter scale

[c.lieftink at nki.nl]

Hi Elin,

Thank you very much for the quick response and elaborate and clear explanation!

Cor

-----Original Message-----
From: Elin [mailto:elin at ebi.ac.uk]
Sent: Wed 9/17/2008 11:08
To: Cor Lieftink
Cc: bioconductor at stat.math.ethz.ch
Subject: Re: [BioC] Definition of the CellHTS2 normalizePlates parameter scale
 
Hi Cor,

I hope I can help you out here. Let me know if it doesn't makes sense.

On the additive scale you look at how much something adds to an effect,
whereas on the multiplicative scale you look for how much something
multiply the effect.

Eg, in a viability screen, if you set the scale to be additive, the
normalized values of the wells will be the difference between each and
your baseline (median, mean, or what every you want). That is wells with
no effect will be more or less zero, whereas wells where cells die or
grow faster will be negative and positive. Using the multiplicative
scale you would instead look at how the growth is multiplied in the
different wells. No effect would be normalized to 1, the number of cells
is 1 times the baseline. A value of 2 would mean that the number of
cells in that well is 2 times the normal, whereas 0.5 would mean that
you have half the cells expected.

So you see, once you have set the scale you have no freedom to choose if
you want to normalize by subtraction or dividing. As this inf  fact is
what decides which scale you have. Something normalized by subtracting
the median *is* on the additive scale and vice versa.

In cellHTS2 the parameter "scale" will control the normalization and
transformations preformed on the data set."Additive" will normalize by
subtracting the median (or what you set the "method" to be).
"Multiplicative" will normalize by dividing with the median. Setting,
scale="multiplicative" and log =TRUE, will log transform your data and
then set the scale to "additive",
(log(toBeNormalized)-log(median)=log(toBeNormalized)/(median)).

The different options are there to cater for different kind of
experimental setups and biological questions. I don't know which scale
is appropriate for your analysis, but if you think the right thing to do
is to normalize by dividing by the median, then you think the scale you
want is multiplicative. Setting scale= "multiplicative", log=FALSE, will
do just what you want (and nothing more :) )

Hope it helps,

Elin


c.lieftink at nki.nl wrote:
> Hello everyone,
>
> The cellHTS2 method normalizePlates has a parameter called "scale". The two values are "additive" vs. "multiplicative".  Although it seems be some basic statistical knowledge I am not able to find a clear definition of additive versus multiplicative data scale. Can anyone provide me with one? I have searched the internet but could not find one.
>
> 2) In cellHTS2 the scale value has the following influence on normalizing with "median":
> additive: the measurement is divided by the median
> multiplicative: the median is subtracted from the measurement
>
> I assume that a cellcount is additive data. It is my understanding that the most used method for normalization in RNAi screens is dividing by the median. However, cellHTS2 does not offer that option for additive data. Would it a good idea when the end user is offered to possibility to choose for his/her (additive) data to subtract or divide by the median?
>
> 3. I am working with cellHTS2 2.2.5. Maybe something has changed in 2.4.1, but I was not able to find any release notes.
>
> Best regards,
>
> Cor
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor
>