Hi Jim Thanks for the continued excellent suggestions. One quick point on the spike-ins. I'm pretty sure this comes pre-mixed nowadays so the ratios are fixed. I'll double check. Your point about the spiking-in being done after RNA extraction is well-taken, but I think the techs quantitate the RNA before labeling, etc. Again, I'll double check and try to get a sense of how accurate people think this. To possibly give you a better sense of what I'm looking at, the attached figure shows the means of the 10 positive E1A controls, plus the 308 negative controls, plus my favorite gene (huntingtin in red) across an 89 array slice of my dataset. This is after bg correction and quantile normalization using limma. More tomorrow. Best, Nat On Feb 4, 2013, at 12:53 PM, James W. MacDonald wrote: > Hi Nat, > > On 2/4/2013 3:31 PM, Nathan (Nat) Goodman wrote: >> Hi Jim >> >> Everything you mentioned is good, and I agree straightforward to program up by hand. The other things I'd like to do are equally obvious and probably not too hard. >> >> 1) Use the negative controls to define the limit of detection. > > See filter.wellaboveNEG in the Agi4x44PreProcess package. > >> >> 2) Use the positive controls to define the standard curve -- aka normalization -- or at least confirm that normalization worked as expected. > > I like the idea of checking things, but am less enthused about using the positive controls to normalize. These are spiked in by shaky- handed technicians, and are done as a different step from extracting total RNA. > > As an ex-lab rat with extensive immunoassay experience I am highly suspect of any serial dilution that involves measuring (and aliquotting) 2?l of a solution using a pipettor. I just don't believe it can be done accurately, and is a recipe for uber high variance for the standard curve. > >> >> 3) Propagate the variance estimates from the replicated probes to downstream tests of significance. > > You won't be able to do that with limma, unless I am missing something. And I don't think that is the right thing to do anyway. The variance estimates you are talking about are intra-array variances, which tend to be smaller than the inter-array variances that the eBayes() step in limma is adjusting for. > > And if you were to propagate the intra-array variances, it would only be reasonable to do so for the replicated spots. But if you are interested in propagating uncertainty, you might look at the puma package. > > Best, > > Jim > > >> >> Before I forget, I want to thank you for taking the time to engage in this conversation. I really appreciate the help. >> >> Best, >> Nat >> >> On Feb 4, 2013, at 11:40 AM, James W. MacDonald wrote: >> >>> I guess it depends on what you want to do with the positive and negative controls and the replicated stuff. I might be lacking vision here, but it seems to me there are only limited things that can be done. The only interesting things I have ever come up with are >>> >>> Boxplots of different types of controls, by array. >>> Scatter plots of the spike-in controls. You could get fancy here and fit linear models and stuff, but I find that sort of boring and uninteresting. I just want to see that they look relatively similar after normalization. >>> Average replicates of non-controls, or maybe better - just use a single observation so you aren't smoothing. >>> >>> I don't use the Agi4x44PreProcess package for any of that, because it is really simple to do by hand. Did you want to do something else? >>> >>> Best, >>> >>> Jim >>> >>> >>> >>> On 2/4/2013 2:26 PM, Nathan (Nat) Goodman wrote: >>>> I've seen Agi4x44PreProcess, too. As far as I can tell, it simply averages the replicas (!!??). I'll look at it more deeply if you think it might do more. >>>> >>>> Best, >>>> Nat >>>> >>>> On Feb 4, 2013, at 11:16 AM, James W. MacDonald wrote: >>>> >>>>> Hi Nat, >>>>> >>>>> The Agi4x44PreProcess package does some things with the controls on the Agilent 4x44 array format, and you might look there for inspiration. >>>>> >>>>> Best, >>>>> >>>>> Jim >>>>> >>>>> On 2/4/2013 2:10 PM, Nathan (Nat) Goodman wrote: >>>>>> Thanks, Jim. I am already using limma which does the basic processing quite well, but I don't think it does anything with the positive and negative controls or the numerous replicated non-control probes on the Agilent array. I'm looking for a package that does something useful with these features. >>>>>> >>>>>> Best, >>>>>> Nat >>>>>> >>>>>> On Feb 4, 2013, at 10:44 AM, James W. MacDonald wrote: >>>>>> >>>>>>> Hi Nat, >>>>>>> >>>>>>> On 2/4/2013 9:47 AM, Nathan (Nat) Goodman wrote: >>>>>>>> Greetings- I have been unable to find a bioc package which does for the agilent Mouse 8x60K array, what the affy package does for affymetrix arrays. Any pointers? >>>>>>> These Agilent arrays have a single 60-mer per transcript, so don't require something like the affy package (which is intended to summarize multiple 25-mers for a transcript to a single statistic). Instead, you most likely just need something like limma, which has the necessary functionality to read the data in, read in the GAL file so you annotate your output, normalize, and make comparisons. >>>>>>> >>>>>>> The limma User's Guide has several Agilent examples, IIRC, so I would start there. >>>>>>> >>>>>>> Best, >>>>>>> >>>>>>> Jim >>>>>>> >>>>>>> >>>>>>>> Many thanks, >>>>>>>> Nat Goodman >>>>>>>> ISB >>>>>>>> >>>>>>>> >>>>>>>> [[alternative HTML version deleted]] >>>>>>>> >>>>>>>> _______________________________________________ >>>>>>>> Bioconductor mailing list >>>>>>>> Bioconductor at r-project.org >>>>>>>> https://stat.ethz.ch/mailman/listinfo/bioconductor >>>>>>>> Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor >>>>>>> -- >>>>>>> James W. MacDonald, M.S. >>>>>>> Biostatistician >>>>>>> University of Washington >>>>>>> Environmental and Occupational Health Sciences >>>>>>> 4225 Roosevelt Way NE, # 100 >>>>>>> Seattle WA 98105-6099 >>>>>>> >>>>> -- >>>>> James W. MacDonald, M.S. >>>>> Biostatistician >>>>> University of Washington >>>>> Environmental and Occupational Health Sciences >>>>> 4225 Roosevelt Way NE, # 100 >>>>> Seattle WA 98105-6099 >>>>> >>> -- >>> James W. MacDonald, M.S. >>> Biostatistician >>> University of Washington >>> Environmental and Occupational Health Sciences >>> 4225 Roosevelt Way NE, # 100 >>> Seattle WA 98105-6099 >>> > > -- > James W. MacDonald, M.S. > Biostatistician > University of Washington > Environmental and Occupational Health Sciences > 4225 Roosevelt Way NE, # 100 > Seattle WA 98105-6099 >