[BioC] questions of using Limma: should I include allthesam p les?

[Wu, Xiwei]

Thanks a lot for your explanation. With the so-called 1x3 design because of
the missing of one treatment group, is it more like a one-way anova now
instead of 2-way anova? If so, does the coefficient still have the same
meaning as in a 2x2 factorial design? What I mean by that is: does C still
refers to the intercept? Does A still be main effect of A? and AB still the
interaction (although it can not be estimated solely based on the design)?

Xiwei

-----Original Message-----
From: Fangxin Hong [mailto:fhong at salk.edu]
Sent: Tuesday, February 08, 2005 11:16 AM
To: Wu, Xiwei
Cc: 'bioconductor at stat.math.ethz.ch'
Subject: RE: [BioC] questions of using Limma: should I include allthesam
ples?



> I am also facing a little bit more complicated situation:
> There are actually three types of inhibitors, which makes it a 2x4 design
> (Am I right?)
Yes.

> If the biological interest is to see each inhibitor effect separately,
> does
> it make sense to include all the chips together for the analysis, or only
> one inhibitor at a time?
As long as the noisy scale are similar on all chips, include all the chips
together for the analysis will give you better power in detecting
differentially expressed genes, without messing up the analysis.




> In addition, what if the inhibitor only experiments are not done? Does it
> kill the analysis totally?
> The design matrix would be sth like this:
> 1 0 0 (only C)
> 1 1 0 (C+A)
> 1 1 1 (C+A+B+AB)
>
> to estimate C, A, and B+AB assuming that B effect is little. My question
> is
> whether the linear model still hold (since B is now included in error
> variance?).
This will only means that the effect of B is undistinguishable from AB
interaction, but liner model will still hold ( treat is as 1*3 design).
Your design matrix is right/ But remember what you get is not inhibitor
only effect.

Fangxin



> -----Original Message-----
> From: Fangxin Hong [mailto:fhong at salk.edu]
> Sent: Tuesday, February 08, 2005 8:58 AM
> To: Wu, Xiwei
> Subject: RE: [BioC] questions of using Limma: should I include all
> thesam ples?
>
>
>
>> Fangxin,
>>
>> Thank you very much for your reply.
>> Sorry the contrast matrix should read:
>> -1  1  0  0
>> 1  -1 -1  1
> This is right for what you want.
>
>
>> The design and contrast matrix do look more clear as you suggested, but
>> if
>> these different matrix were used, would the result be different at all?
> No, there should not be any difference in the result you get.
>
> Fangxin
>
>> Xiwei
>>
>> -----Original Message-----
>> From: Fangxin Hong [mailto:fhong at salk.edu]
>> Sent: Monday, February 07, 2005 4:58 PM
>> To: Wu, Xiwei
>> Cc: 'bioconductor at stat.math.ethz.ch'
>> Subject: Re: [BioC] questions of using Limma: should I include all the
>> samples?
>>
>>
>>
>>> I am trying to use Limma with design matrix of
>>>
>>>                 1 0 0 0
>>>                 1 0 0 0
>>>                 1 0 0 0
>>>                 0 1 0 0
>>>                 0 1 0 0
>>>                 0 1 0 0
>>>                 0 0 1 0
>>>                 0 0 1 0
>>>                 0 0 1 0
>>>                 0 0 0 1
>>>                 0 0 0 1
>>>                 0 0 0 1
>>>
>>> to estimate the four coefficinet of C, C+ A, C+B and C+A+B+AB (of
>> course,
>>> I
>>> can estimate A, B, and AB directly using a different design matrix).
>>>
>>> Since the contrast of interest is A and AB, so the contrast matrix
>> should
>>> be:
>>> -1  1  0  0
>>> -1 -1 -1  1
>>>
>>> My question is:
>>> 1) Are the design and contrast matrix correct?
>> If your design matrix is right, then your contrast marix is not right,
>> as
>> the (-1,-1,-1,1) will give you estimate of AB-2C, but not AB.
>>
>> I would suggest you estimate C, A, B, and AB
>> using design matrix
>> 1 0 0 0 (only C)
>> 1 1 0 0(C+A)
>> 1 0 1 0(C+B)
>> 1 1 1 1 (C+A+B+AB)
>>
>> and construct your contrast as
>> 0 1 0 0 (test A)
>> 0 0 0 1  (test AB)
>>
>>
>>
>>> 2) I know this is a very naive question, but if I am only interested in
>> hormone only effect, can I just use the untreated and hormone alone
>> treated
>>> samples as the input (so instead of the 12 CEL files, only use the
>>> first
>> 6
>>> CEL files)? Will the analysis result be the same or different if not
>> counting the normalization-produced difference? If there is difference,
>> is
>>> that due to the difference of df?
>> Well, this will only affect your error variance estimation, since you
>> lose
>> power for it. Usually less genes will be identified out using subset of
>> the data, is indeed you can assume one model for all 12 data sets.
>>
>> Hopefull this would help.
>>
>> Fangxin
>>
>>
>>
>>
>> --
>> Fangxin Hong, Ph.D.
>> Plant Biology Laboratory
>> The Salk Institute
>> 10010 N. Torrey Pines Rd.
>> La Jolla, CA 92037
>> E-mail: fhong at salk.edu
>>
>>
>>
>>
>>
>>
>>
>> -----------------------------------------------------------
>> SECURITY/CONFIDENTIALITY WARNING:  This message and any attachments are
>> intended solely for the individual or entity to which they are
>> addressed.
>> This communication may contain information that is privileged,
>> confidential, or exempt from disclosure under applicable law (e.g.,
>> personal health information, research data, financial information).
>> Because this e-mail has been sent without encryption, individuals other
>> than the intended recipient may be able to view the information, forward
>> it to others or tamper with the information without the knowledge or
>> consent of the sender. If you are not the intended recipient, or the
>> employee or person responsible for delivering the message to the
>> intended
>> recipient, any dissemination, distribution or copying of the
>> communication
>> is strictly prohibited. If you received the communication in error,
>> please
>> notify the sender immediately by replying to this message and deleting
>> the
>> message and any accompanying files from your system. If, due t!
>>  o the
>> security risks, you do not wish to receive further communications via
>> e-mail, please reply to this message and inform the sender that you do
>> not
>> wish to receive further e-mail from the sender.
>> ===========================================================
>
>
> --
> Fangxin Hong, Ph.D.
> Plant Biology Laboratory
> The Salk Institute
> 10010 N. Torrey Pines Rd.
> La Jolla, CA 92037
> E-mail: fhong at salk.edu
>
>


-- 
Fangxin Hong, Ph.D.
Plant Biology Laboratory
The Salk Institute
10010 N. Torrey Pines Rd.
La Jolla, CA 92037
E-mail: fhong at salk.edu


	[[alternative HTML version deleted]]