> 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@stat.math.ethz.ch > [mailto:bioconductor-bounces@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@wehi.edu.au] >>Enviada em: ter?a-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@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 / Vers?o: 1.3.13 > Proteja o seu e-mail Terra: http://www.emailprotegido.terra.com.br/