[BioC] Confusion over limma documentation and design/contrast

michael watson (IAH-C) michael.watson at bbsrc.ac.uk
Fri Nov 12 14:42:22 CET 2004


Hi Gordon

Thanks for the response.  I worked through my two examples and found out
they are equivalent.  

>Do you know how to multiply a vector by a matrix? If you do, then I
think 
>the best way to figure out what the design matrix is doing for you is
just 
>to sit down with a piece of paper and a pencil for a few minutes, and 
>multiply the design matrix by the coefficient vector.

I did actually do this previous to sending the mail but it didn't easy
my confusion.  Eg:

> apoai.design
   ApoAI Control
1      0       1
2      0       1
3      0       1
4      0       1
5      0       1
6      0       1
7      0       1
8      0       1
9      1       0
10     1       0
11     1       0
12     1       0
13     1       0
14     1       0
15     1       0
16     1       0

> apoai.contrasts
        ApoAI - Control
ApoAI                 1
Control              -1

> apoai.design %*% apoai.contrasts
    
     ApoAI - Control
  1               -1
  2               -1
  3               -1
  4               -1
  5               -1
  6               -1
  7               -1
  8               -1
  9                1
  10               1
  11               1
  12               1
  13               1
  14               1
  15               1
  16               1

So I still don't see how multiplying those two matrices relates to the
ApoAI design matrix given in the documentation, which is:

    Control-Ref KO-Control
c1           1          0
c2           1          0
c3           1          0
c4           1          0
c5           1          0
c6           1          0
c7           1          0
c8           1          0
k1           1          1
k2           1          1
k3           1          1
k4           1          1
k5           1          1
k6           1          1
k7           1          1
k8           1          1

I have read the updated limma documentation and it is clearer now, but
it is still unclear why the above design matrix is used for the ApoAI
data - there is no explanation as to how it relates to the design matrix
and contrasts matrix I calculated above using modelMatrix() and
makeContrasts().

Mick

-----Original Message-----
From: Gordon Smyth [mailto:smyth at wehi.edu.au] 
Sent: 12 November 2004 12:23
To: michael watson (IAH-C)
Cc: bioconductor at stat.math.ethz.ch
Subject: Re: Bioconductor Digest, Vol 21, Issue 12


You are reading the Limma User's Guide written for Bioconductor Release 
1.4. Could I encourage you to consider the moving onto Bioconductor 1.5?

Even without upgrading your software, you could read a more recent
User's 
Guide at http://bioinf.wehi.edu.au/limma/usersguide.pdf. Section 10
might 
help, especially section 10.5.

>Date: Thu, 11 Nov 2004 12:11:34 -0000
>From: "michael watson (IAH-C)" <michael.watson at bbsrc.ac.uk>
>Subject: [BioC] Confusion over limma documentation and design/contrast
>         matrices
>To: <bioconductor at stat.math.ethz.ch>
>
>Hi
>
>The confusion comes from the examples given in usersguide.pdf.  
>Sections 7.3 and 8.2 both deal with two colour microarray data where a 
>common reference has been used.
>
>Section 7.3 advocates the use of designMatrix() and makeContrasts(), 
>where lmFit() is first used with the design matrix, and then
>contrasts.fit() is used with the contrasts matrix, and then eBayes() 
>applied to the resulting linear model fit.
>
>Section 8.2 gives us the design matrix directly, and does not use a 
>contrasts matrix at all.
>
>So my first question is this: with two colour cDNA microarrays which 
>use a common reference, do we need both a design matrix and a contrasts

>matrix, or just a design matrix?  (I'm assuming it depends on how many 
>samples there are in addition to the reference, but I'm not sure).
>
>Looking at section 8.2, the ApoAI data, we have three samples - control

>mice, ApoAI mice and the reference sample.  8 control mice were 
>compared to the reference and 8 ApoAI mice were compared to the 
>reference, and in all cases the reference was Cy3.  Therefore, a 
>targets file to describe this experiment can reasonably be expected to 
>look like this:
>
>    SlideNumber     Cy5 Cy3
>1           c1 Control Ref
>2           c2 Control Ref
>3           c3 Control Ref
>4           c4 Control Ref
>5           c5 Control Ref
>6           c6 Control Ref
>7           c7 Control Ref
>8           c8 Control Ref
>9           k1   ApoAI Ref
>10          k2   ApoAI Ref
>11          k3   ApoAI Ref
>12          k4   ApoAI Ref
>13          k5   ApoAI Ref
>14          k6   ApoAI Ref
>15          k7   ApoAI Ref
>16          k8   ApoAI Ref
>
>If we run modelMatrix() on this, we get this:
>
>    ApoAI Control
>1      0       1
>2      0       1
>3      0       1
>4      0       1
>5      0       1
>6      0       1
>7      0       1
>8      0       1
>9      1       0
>10     1       0
>11     1       0
>12     1       0
>13     1       0
>14     1       0
>15     1       0
>16     1       0
>
>If we then carry on the example in 7.3, and create a contrasts matrix, 
>then I assume we are interested in the comparison of ApoAI knockout 
>mice to Control mice, so I run:
>
> > makeContrasts(ApoAI-Control, levels=apoai.design)
>
>And get this
>
>         ApoAI - Control
>ApoAI                 1
>Control              -1
>
>So, just to take a breath here, I am using the documentation from 
>section 7.3 to run against the data in section 8.2.  I now have a 
>design matrix and a contrasts matrix for the ApoAI data.  The only 
>problem is that they look completely different to the example design 
>matrix given in section 8.2.  The design matrix there is:
>
>    Control-Ref KO-Control
>c1           1          0
>c2           1          0
>c3           1          0
>c4           1          0
>c5           1          0
>c6           1          0
>c7           1          0
>c8           1          0
>k1           1          1
>k2           1          1
>k3           1          1
>k4           1          1
>k5           1          1
>k6           1          1
>k7           1          1
>k8           1          1
>
>And there is no contrasts matrix.
>
>So, my question 2: is the design matrix given in section 8.2 the 
>equivalent of using the design and contrasts matrices that I calculated

>above?

Yes it is. This has been said many, many times on this list.

Do you know how to multiply a vector by a matrix? If you do, then I
think 
the best way to figure out what the design matrix is doing for you is
just 
to sit down with a piece of paper and a pencil for a few minutes, and 
multiply the design matrix by the coefficient vector.

Gordon

>Question 3: does the fact that the ApoAI data only has one contrast of 
>interest mean that I don't need a contrasts matrix, I only need a 
>design matrix? (albeit a DIFFERENT design matrix to one that would be 
>given by modelMatrix()).
>
>Question 4: can anyone recommend a good book which I can go and read 
>and learn all about design and contrast matrices and their relation to 
>linear models?  Everything I have read so far just says something like 
>"... The design matrix is therefore this... and the contrast matrix is 
>therefore this..." and doesn't actually explain how exactly the 
>structure of the matrices was arrived at...
>
>Thanks and sorry for the long mail!
>
>Mick



More information about the Bioconductor mailing list