Hi Josep, > Dear Heidi, > > I am using your package HTqPCR, in order to analyze miRNA taqman plates, > for > which this package is extremely useful. > > For normalization, I used "rank invariant" which selects two miRNAs as the > most "rank invariant". However, as a "sanity check", I wanted to check if > the results normalizing by the "deltaCt" method using the same miRNAs as > contorls. > > To my surprise, the normalized values when normalizing by these two miRNAs > using the deltaCt method and the "rank invariant" showed very weak > correlation. I was expecting that the correlations should be similar. > Should > I worry about that? What is the method I should trust most, then? > Although using the same genes for normalisation, these two methods don't necessarily give the same results at all. For deltaCt, the average Ct value of the genes is calculated, and this is simply subtracted from all other Ct values. So this is a strictly linear normalisation method. If you plot the raw versus the normalised value, you'll notice that these are simply shifted compared to the diagonal. The rank invariant method on the other hand is a non-linear normalisation method. The rank invariant genes are identified, a smoothed line is fitted though the corresponding Ct values, and this smoothing is subsequently applied to all the Ct values on the plate. Which method to go for depends on your data. deltaCt is the "classical" method that is commonly used for mRNA genes using e.g. actin and GAPDH. However, I personally don't always like it 100%. These reference genes tend to be expressed at very high levels, i.e. have low Ct values. Therefore, they might not be representative of any biases that might be present for genes expressed at lower levels, i.e. at the high end of the Ct range. A smoothing approach can often be more robust, at least if the rank invariant genes cover a broad range of the Ct values observed in your data. In your case I notice that there are only 2 rank invariant genes though. This might still be okay, but if you have more than a couple of samples, you might try decreasing hte scale.rank.samples parameter in normalizeCtData(). This indicates how many samples each gene has to be rank invariant across in order to be used for the normalisation. Which a high number of samples, this can be relaxed a bit. Please also note that for miRNAs that have a been a couple of publications suggesting simply normalising by subtracting the mean Ct of all miRNAs present on the plate. This option isn't included in the release version of HTqPCR, however I have the snippet of code for it, and can send it to you if you're interested. Alternatively, selecting norm="deltaCt" and setting deltaCt.genes equal to all your featureNames should in principle give you the same result. HTH \Heidi > Find attached the data and the code used as an example. > > I would be grateful if you could give me a reply regarding this issue. > > Thank you for your time and consideration. > > Best regards, > > Josep > -- > Josep Maria Mercader, PhD > Barcelona Supercomputing Center > Life Science Department, Computational Genomics Group. > josep.mercader at bsc.es > josepmaria.mercader at gmail.com > www.bsc.es >

Hi Josep > Dear Heidi, > > Thank you for your explanations. I think that your suggestions will be > very > helpful, and I am sure that you solved all my uncertainties. I would be > grateful if you could send me this sniped code for normalizing with the > mean > of all miRNAs. > You can give this a go. It's a linear scaling, and assumes that your total amount of miRNA expression is constant. It doesn't required you to define any house-keeping genes though, which can be notoriously tricky for miRNAs. # Load test data data(qPCRraw) # Define plotting function (or just use the individual commands directly) my.norm.method <- function(q, Ct.max=35, ref=1) { # Get the data data <- exprs(q) # For each column, calculate the geometric mean of Ct values<ct.max geo.mean="" <-="" apply(data,="" 2,="" function(x)="" {="" xx="" <-="" log2(subset(x,="" x<ct.max))="" 2^mean(xx)})="" #="" calculate="" the="" scaling="" factor="" geo.scale="" <-="" geo.mean="" geo.mean[ref]="" #="" adjust="" the="" data="" accordingly="" data.norm="" <-="" t(t(data)="" *="" geo.scale)="" #="" return="" the="" normalised="" object="" exprs(q)="" <-="" data.norm="" q="" }="" #="" normalise="" q.norm="" <-="" my.norm.method(qpcrraw)="" #="" plot="" raw="" versus="" normalised="" data="" plot(exprs(qpcrraw),="" exprs(q.norm),="" col="rep(1:n.samples(q.norm)," each="n.wells(q.norm)))"> Besides that, I have yet another question. I am planning to validate some > of > the miRNAs for which I found interesing results, but not using > microfluidic > plates. Is your program also available to work with data of few miRNAs > (let's say 10 or 20), using only the deltaCt method, of course? Is there a > function to easily load a dataframe of qPCR experiments and convert them > to > a qPCR object? > Sure. You can read in you data using any of the "standard" functions, and then use new() to generate a qPCRset object. For example: > # Generate some test data > temp<- matrix(1, nrow=10, ncol=4) > # In your case, this would probably be read in using read.table() > > # Add some row/column names > rownames(temp) <- paste("gene", 1:10, sep="") > colnames(temp) <- paste("sample", 1:4, sep="") > # Usually, row/column names will be present in you data, so these two lines can be ignored > > # Create qPCRset > q <- new("qPCRset", exprs=temp, sampleNames=colnames(temp), featureNames=rownames(temp), featureCategory=as.data.frame(array("OK", dim=dim(temp)))) > q An object of class "qPCRset" Size: 10 features, 4 samples Feature types: Feature names: gene1 gene2 gene3 ... Feature classes: Feature categories: OK Sample names: sample1 sample2 sample3 ... HTH \Heidi > Thank you for all your help and your time again. > > Best regards, > > Josep > > On Sat, Jun 11, 2011 at 8:19 AM, Heidi Dvinge <heidi at="" ebi.ac.uk=""> wrote: > >> Hi Josep, >> >> > Dear Heidi, >> > >> > I am using your package HTqPCR, in order to analyze miRNA taqman >> plates, >> > for >> > which this package is extremely useful. >> > >> > For normalization, I used "rank invariant" which selects two miRNAs as >> the >> > most "rank invariant". However, as a "sanity check", I wanted to check >> if >> > the results normalizing by the "deltaCt" method using the same miRNAs >> as >> > contorls. >> > >> > To my surprise, the normalized values when normalizing by these two >> miRNAs >> > using the deltaCt method and the "rank invariant" showed very weak >> > correlation. I was expecting that the correlations should be similar. >> > Should >> > I worry about that? What is the method I should trust most, then? >> > >> Although using the same genes for normalisation, these two methods don't >> necessarily give the same results at all. >> >> For deltaCt, the average Ct value of the genes is calculated, and this >> is >> simply subtracted from all other Ct values. So this is a strictly linear >> normalisation method. If you plot the raw versus the normalised value, >> you'll notice that these are simply shifted compared to the diagonal. >> >> The rank invariant method on the other hand is a non-linear >> normalisation >> method. The rank invariant genes are identified, a smoothed line is >> fitted >> though the corresponding Ct values, and this smoothing is subsequently >> applied to all the Ct values on the plate. >> >> Which method to go for depends on your data. deltaCt is the "classical" >> method that is commonly used for mRNA genes using e.g. actin and GAPDH. >> However, I personally don't always like it 100%. These reference genes >> tend to be expressed at very high levels, i.e. have low Ct values. >> Therefore, they might not be representative of any biases that might be >> present for genes expressed at lower levels, i.e. at the high end of the >> Ct range. >> >> A smoothing approach can often be more robust, at least if the rank >> invariant genes cover a broad range of the Ct values observed in your >> data. In your case I notice that there are only 2 rank invariant genes >> though. This might still be okay, but if you have more than a couple of >> samples, you might try decreasing hte scale.rank.samples parameter in >> normalizeCtData(). This indicates how many samples each gene has to be >> rank invariant across in order to be used for the normalisation. Which a >> high number of samples, this can be relaxed a bit. >> >> Please also note that for miRNAs that have a been a couple of >> publications >> suggesting simply normalising by subtracting the mean Ct of all miRNAs >> present on the plate. This option isn't included in the release version >> of >> HTqPCR, however I have the snippet of code for it, and can send it to >> you >> if you're interested. Alternatively, selecting norm="deltaCt" and >> setting >> deltaCt.genes equal to all your featureNames should in principle give >> you >> the same result. >> >> HTH >> \Heidi >> >> > Find attached the data and the code used as an example. >> > >> > I would be grateful if you could give me a reply regarding this issue. >> > >> > Thank you for your time and consideration. >> > >> > Best regards, >> > >> > Josep >> > -- >> > Josep Maria Mercader, PhD >> > Barcelona Supercomputing Center >> > Life Science Department, Computational Genomics Group. >> > josep.mercader at bsc.es >> > josepmaria.mercader at gmail.com >> > www.bsc.es >> > >> >> >> > > > -- > Josep Maria Mercader, PhD > Barcelona Supercomputing Center > Life Science Department, Computational Genomics Group. > josep.mercader at bsc.es > josepmaria.mercader at gmail.com > www.bsc.es >