[BioC] LIMMA: MA, design, and contrasts

Jenny Drnevich drnevich at uiuc.edu
Thu Sep 20 23:27:31 CEST 2007

Hi Tiandao,

It sounds like you are combining arrays from different experiments - 
one a reference design, another a loop design, and a third that's a 
different loop design. In general, it's NOT good to combine arrays 
from such vastly different experiments, because there will be batch 
effects between the experiments and the differences in the type of 
design (ref. vs. loop) necessitate different analysis methods. You 
should probably keep arrays from each experiment completely separate, 
from the normalization and pre-processing to the statistical 
analysis. If there is a compelling reason to try to combine these 
arrays, then I suggest you find a local statistician or experienced 
microarray analyst to help you, because the chances of messing up 
somewhere along the line are very high!!!


At 04:00 PM 9/20/2007, Tiandao Li wrote:
>Hello Jenny,
>I created a composite design matrix with some common reference arrays and
>arrays from some loop designs, such as loop design 1 and loop design 2. I
>have a big MA with all normalized arrays, and a big contrast matrix to
>find the differentially expressed genes in different treatments only
>existed in loop design 1. Or, I can build a small MA, a small design
>matrix using the same reference, and contrast matrix for some contrasts
>only for arrays in loop design 1. Which way can I get more reliable
>result? Or I will get the same result.
>On Thu, 20 Sep 2007, Jenny Drnevich wrote:
>Hi Tiandao,
>A quick answer to your first question:
> >corfit <- duplicateCorrelation(MA,design,ndups=4) # A slow computation!
>The order of the arrays in the columns of MA **MUST** match the order
>of the arrays in the rows of design, else your design matrix is not
>correct for your MA object.
> >2. I have one MA file including all my experiments, and also an all-in-one
> >contrast matrix including different contrasts (related or un-related).
> >Should I use this all-in-one contrast matrix for linear model to find the
> >differentially expressed genes? This doesn't sound right. Or I use subset
> >of MA for and only for related one or more contrasts, and use all-in-one
> >MA only necessary. Which one is better?
>I don't quite understand your second question... each contrast in
>your contrast matrix will be estimated separately using only those
>columns of MA that are indicated from the design matrix and the
>contrast matrix. Perhaps if you could explain your question in more
>detail with example code, we could better answer it.
> >Thanks in advance,
> >
> >Tiandao
> >
> >_______________________________________________
> >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
>Jenny Drnevich, Ph.D.
>Functional Genomics Bioinformatics Specialist
>W.M. Keck Center for Comparative and Functional Genomics
>Roy J. Carver Biotechnology Center
>University of Illinois, Urbana-Champaign
>330 ERML
>1201 W. Gregory Dr.
>Urbana, IL 61801
>ph: 217-244-7355
>fax: 217-265-5066
>e-mail: drnevich at uiuc.edu
>Bioconductor mailing list
>Bioconductor at stat.math.ethz.ch
>Search the archives: 

Jenny Drnevich, Ph.D.

Functional Genomics Bioinformatics Specialist
W.M. Keck Center for Comparative and Functional Genomics
Roy J. Carver Biotechnology Center
University of Illinois, Urbana-Champaign

330 ERML
1201 W. Gregory Dr.
Urbana, IL 61801

ph: 217-244-7355
fax: 217-265-5066
e-mail: drnevich at uiuc.edu

More information about the Bioconductor mailing list