RES: [BioC] Genechip experiment in limma

Gordon K Smyth smyth at wehi.EDU.AU
Sun Aug 24 03:14:34 MEST 2003


> My spatial heterogeneity is intra-print-tip. The unit must be smaller
> that the sector.
>
> Your assumption about my Genechip hybridization classification is
> correct.
> (C01h, C02h, C04h, C06h, A01h, A02h, A04h, A06h, B01h, B02h, B04h,
> B06h).
>
> I want to know what is differentially expressed between Trt A x Ctrl,
> Trt B x Ctrl, Trt x Ctrl and Trt A x Trt B.
>
>>Yes, there is a design matrix which incorporates the temporal
> information,
>>but it is difficult to do statistical analysis in the absence of any
>> replication. (I do wish people would be prepared to invest in an extra
> two
>>or three chips in order to be able to do statistical analysis.)
>
> The absence of replication is not caused by money but by lack of
> biological material.
> I really want to evaluate the DE with a homogeneous pool of starting
> cells. In this case, there was material for just 12 hybridizations. In
> my sense, when doing time series experiments, and you have a limited
> amount of material, it is better to increase the number of time points
> instead of the number of replicates by time points. And the temporal
> information can be used as "replication" (the 4 time points for each
> sample are from the same cell pool, but they have different fates since
> the beggining of the experiment. I do not take aliquots from the same
> cell culture at 1h, 2h, 4h and 6h). I know they are not biological
> replicates in a strict sense, but my study objectives and implications
> are "aware" of this.
>
> So, my question is:
>
> May I create a design and contrast matrix that use the time points as
> replicates, but in a more flexible way. For instance, taking the first
> three time points as replicates of a early process, and the last three
> time points as replicates of a late process, without doing three
> diferent analysis?

Well, you would have to define two different design matrices and do two
separate fits.

Gordon

> Christian
>
> -----Mensagem original-----
> De: bioconductor-bounces at stat.math.ethz.ch
> [mailto:bioconductor-bounces at stat.math.ethz.ch] Em nome de Gordon Smyth
> Enviada em: sexta-feira, 22 de agosto de 2003 03:05
> Para: Christian Probst
> Cc: BioC Mailing List
> Assunto: [BioC] Genechip experiment in limma
>
> At 02:48 AM 22/08/2003, Christian Probst wrote:
>>Gordon,
>>
>>I succeed in loading the cDNA microarray data. As I need a spatial
>> normalization, I will try to do it before the use of limma package.
>
> Remember that print-tip-loess is in itself a simple form of spatial
> normalization (as well as intensity-based normalization). In my
> experience,
> print-tip-loess does most of what spatial normalization would have done.
>
> But, as you say, if you really do want to do formal 2D normalization
> you'll
> have to do it outside of limma.
>
>>I am also running a Genechip experiment, with four time points and
> three
>>treatments (A,B and Ctrl), with no replicates.
>
> Am I correct in assuming that your first 4 chips are hybridised with
> Ctrl,
> the second four with A and the last four with B? Are the time points
> 1,2,3,4,1,2,3,4,1,2,3,4?
>
>>I would like to ask you about the design table in order to analyse this
>> data.
>>I want to find those genes that are differentially expressed in AxCtrl,
>> BxCtrl, and AB x Ctrl.
>
> Do you want genes which are DE between A and Ctrl and DE between B and
> Ctrl? Obviously this includes identifying genes which are both DE for A
> vs
> Ctrl and B vs Ctrl. I am not sure what you mean by "AB x Ctrl", perhaps
> you
> just mean different for both A vs Ctrl and B vs Ctrl.
>
>>Is this design matrix correct?
>
> No it's not. If my assumptions above are correct, and your time points
> were
> really replicates, which they're not, then you'd want either
>
>  > design
> Ctrl A B
> 1 0 0
> 1 0 0
> 1 0 0
> 1 0 0
> 0 1 0
> 0 1 0
> 0 1 0
> 0 1 0
> 0 0 1
> 0 0 1
> 0 0 1
> 0 0 1
>
> with contrast matrix
>
>  > cont.matrix
> A-Ctrl  B-Ctrl
> -1    -1
>   1     0
>   0     1
>
>  > design
> Ctrl A-Ctrl B-Ctrl
> 1 0 0
> 1 0 0
> 1 0 0
> 1 0 0
> 1 1 0
> 1 1 0
> 1 1 0
> 1 1 0
> 1 0 1
> 1 0 1
> 1 0 1
> 1 0 1
>
> with contrast matrix
>
>  > cont.matrix
> A-Ctrl  B-Ctrl
> 0     0
> 1     0
> 0     1
>
> I suggest using the first design matrix, in which case using limma
> 1.1.11,
> you would go
>
> fit <- lmFit(eset, design)
> fit <- contrasts.fit(fit, cont.matrix)
> fit <- eBayes(fit)
> clas <- classifyTests(fit)
> vennDiagram(clas)
>
> to classify each gene as differentially expressed or not for AvsCtrl,
> BvsCtrl or both.
>
>> >design
>>      A  B AB
>>1    0  0  0
>>2    0  0  0
>>3    0  0  0
>>4    0  0  0
>>5    1  0  1
>>6    1  0  1
>>7    1  0  1
>>8    1  0  1
>>9    0  1  1
>>10   0  1  1
>>11   0  1  1
>>12   0  1  1
>>
>>As I have a time-course experiment, and I am using the four time points
>> as replicates, in order to analyze this data statistically, I wonder if
>> there is a design table that uses the temporal information in a
> flexible
>>way, to identify DEG in the beggining or ending of the infection
>>proccess (I will mix my "replication" with my "temporal framework").
>
> Yes, there is a design matrix which incorporates the temporal
> information,
> but it is difficult to do statistical analysis in the absence of any
> replication. (I do wish people would be prepared to invest in an extra
> two
> or three chips in order to be able to do statistical analysis.)
>
> There are things you could do, for example using genes which are not
> differentially expressed to estimate variability for genes which are
> differentially expressed, but this is a research problem.
>
> Regards
> Gordon
>
>>TIA
>>
>>Christian
>>
>>
>>-----Mensagem original-----
>>De: Gordon Smyth [mailto:smyth at wehi.edu.au]
>>Enviada em: terça-feira, 19 de agosto de 2003 23:01
>>Para: Christian M. Probst
>>Assunto: Re: Limma error
>>
>>Christian,
>>
>>At 02:50 AM 20/08/2003, Christian M. Probst wrote:
>> >Dear Mr. Smyth,
>> >
>> >I am trying to use the limma package, but in the first step of the
> data
>> >loading, I run into the following error:
>> >
>> > > library(limma)
>> > > files<-dir(pattern="*\\.spot")
>> > > files
>> >[1] "EP3_05_ET5_05_0808.spot" "EP3_28_ET5_28_0908.spot"
>> >[3] "SP3_30_EP5_28_0908.spot" "SP3_30_ST5_30_0908.spot"
>> > > RG<-read.maimages(files,source="spot")
>> >Read EP3_05_ET5_05_0808.spot
>> >Read EP3_28_ET5_28_0908.spot
>> >Error in "[<-"(*tmp*, , i, value = NULL) :
>> >         number of items to replace is not a multiple of replacement
>>length
>> > >
>> >
>> >This message appears consistently in other files, also. I havent
> found
>>a
>> >description in the Bioconductor discussion list, so I choose to
> bother
>>you
>> >directly.
>>
>>It is pretty hard to give you a lot of help from the above information.
>> It
>>is not true to say that read.maimages consistently gives this error.
> The
>>
>>above output shows that read.maimages read your first two files
>>"EP3_05_ET5_05_0808.spot" and "EP3_28_ET5_28_0908.spot" successfully
> but
>>
>>then fails on "SP3_30_EP5_28_0908.spot". What is different about this
>> third
>>file compared the first two?
>>
>>Also, what version of limma are you using?
>>
>>Gordon
>>
>>
>>
>> >TIA
>> >
>> >Christian
>
> _______________________________________________
> Bioconductor mailing list
> Bioconductor at stat.math.ethz.ch
> https://www.stat.math.ethz.ch/mailman/listinfo/bioconductor
>
> Esta mensagem foi verificada pelo E-mail Protegido Terra.
> Scan engine: VirusScan / Atualizado em 20/08/2003 / Versão: 1.3.13
> Proteja o seu e-mail Terra: http://www.emailprotegido.terra.com.br/



More information about the Bioconductor mailing list