anyone? I found that the GeneSelector package has some nice stuff for leave-1- out (and more) cross validation mark On 28/09/2009, at 11:56 AM, Mark Cowley wrote: > Dear list, > Rather than reinventing the wheel, I hope someone in the community can > point me in the right direction. > > In a pilot study, we have looked at a single xenograft over time > (0,8,24,48 hours) with or without drug treatment (7 groups). Using > Illumina WG6 arrays, with biological quadruplicates and controls at > each time point we have a phenomenal amount of clean signal at 8hr > (3200 genes q < 0.05), which tapers off at 24 and 48hr (307 and 65 > genes q < 0.05 respectively). > This study is about to grow with many more xenograft lines and more > drugs, so from this pilot study data, we'd like to work out (1) how > few replicates we can use, yet still get strong signal, and (2) > whether we need the matched timepoint controls. > > part (1) sounds like a leave-one-out cross validation problem, but i'd > appreciate anyone else's advice here. I've come across the MCREstimate > package on BioC, and was wondering if there were any others worth > using? > part (2) I can probably assess by comparing DE genes treated vs > untreated at each timepoint to the values I get comparing treated at > each timepoint to the time 0 control. It already looks like the > timepoint controls are very similar to the t0, but there must be a > more elegant way to quantify this. > > Here's where I started re-inventing the wheel: > by randomly selecting 3 arrays per group, repeating the limma > analysis, and assessing the number of DE genes that remain, I recover > ~ 78% of genes with q<0.05, but with just 2 replicates I recover only > 30% of the genes. I expect that the loss of statistical power, would > be offset somewhat by the benefits due to moderated t-stats and eBayes > (since I have 7 groups in the model). > > looking for any advice > > cheers, > Mark > > > ---------------------------------------------------------------------- > Mark Cowley, PhD > > Peter Wills Bioinformatics Centre > Garvan Institute of Medical Research > 384 Victoria St | Darlinghurst | NSW 2010 | Australia > phone: +61 2 9295 8542 | fax: +61 2 9295 8538 | email: m.cowley at garvan.org.au > | web: www.garvan.org.au > ---------------------------------------------------------------------- > > > [[alternative HTML version deleted]] > > _______________________________________________ > Bioconductor mailing list > Bioconductor at stat.math.ethz.ch > https://stat.ethz.ch/mailman/listinfo/bioconductor > Search the archives: http://news.gmane.org/gmane.science.biology.informatics.conductor