log transform in RMA normalization

0

Entering edit mode

Jack Luo
▴
440

@jack-luo-4241
Last seen 10.0 years ago

I am trying to understand the details of the RMA algorithm in terms of
log
transform. The probe level data obtained from ReadAffy is obviously
not log
transformed, so in the 3 steps:
1. background subtraction:
2. quantile normalization:
3. median polish summarization.
should the algorithm work on raw data (without log transform) or log
transformed data for these steps? I was trying to judge from the
output
data, but realize that it is not obvious to figure out because the
following two options both give log transformed output:
(take quantile normalization as example)
A. quantile normalization on raw data and then log transform
B. quantile normalization on log transformed data
Thanks,
-Jack
[[alternative HTML version deleted]]

2

Entering edit mode

RMA is a three step algorithm. It is using "background noise removal" which is in fact some convolution algorithm that produce lower values than signal values itself. Then it does quantile normalization and log2 transformation. You can then subtract two values (from two arrays) and obtain a guess of the log fold change.

If you somehow shuffle three steps of RMA, you get slightly different results. Commercial softwares often implements RMA directly using code of R. Irizarry from Bioconductor. Thus the implementations of RMA should be the same and they probably do not shuffle anything.

RMA algorithm is defined as a 3-step process of:

- background correction
- quantile normalisation
- log2 transformation

If you drift from this, then it is not RMA.

0

Entering edit mode

Ben Bolstad
★
1.2k

@ben-bolstad-1494
Last seen 7.1 years ago

the log transformation occurs at the summarization step. i.e. both
background correction and quantile normalization occur on the natural
scale.
Ben
On Jan 24, 2013, at 6:43 AM, Jack Luo <jluo.rhelp at="" gmail.com=""> wrote:
> I am trying to understand the details of the RMA algorithm in terms
of log
> transform. The probe level data obtained from ReadAffy is obviously
not log
> transformed, so in the 3 steps:
>
> 1. background subtraction:
> 2. quantile normalization:
> 3. median polish summarization.
>
> should the algorithm work on raw data (without log transform) or log
> transformed data for these steps? I was trying to judge from the
output
> data, but realize that it is not obvious to figure out because the
> following two options both give log transformed output:
> (take quantile normalization as example)
>
> A. quantile normalization on raw data and then log transform
> B. quantile normalization on log transformed data
>
> Thanks,
>
> -Jack
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at r-project.org
> https://stat.ethz.ch/mailman/listinfo/bioconductor
> Search the archives:
http://news.gmane.org/gmane.science.biology.informatics.conductor

Similar Posts

Loading Similar Posts

Traffic: 475 users visited in the last hour

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