I've got a quick newbie question-- if I'm trying to do two-way ANOVA on a microarray experiment (with a few missing spots) which packages should I install? Can Bioconductor do empirical null distributions for the ANOVA F statistic? Thank you. -------------------------------------- Protect yourself from spam, use http://sneakemail.com

"Alex F. Bokov" <yjih74b02@sneakemail.com> writes: > I've got a quick newbie question-- if I'm trying to do two-way ANOVA > on a microarray experiment (with a few missing spots) which packages > should I install? It would be good to be clear -- ANOVA at the gene level, or are you thinking of mixing in experiment-level parameters? For the former, Biobase would suffice, though other packages (limma, nlme) might be useful as well. This assumes that you've already got normalized data (or else you'd need the appropriate packages) > Can Bioconductor do empirical null distributions for the ANOVA F statistic? R can. See "?pf" for a function computing the distribution of the F statistic. best, -tony --- A.J. Rossini / rossini@u.washington.edu / rossini@scharp.org Biomedical/Health Informatics and Biostatistics, University of Washington. Biostatistics, HVTN/SCHARP, Fred Hutchinson Cancer Research Center. FHCRC: 206-667-7025 (fax=4812)|Voicemail is pretty sketchy/use Email CONFIDENTIALITY NOTICE: This e-mail message and any attachments ... {{dropped}}

Wow, that was fast! Thanks for your answer. rossini-at-blindglobe.net (A.J. Rossini) |bioconductor.org| wrote: > It would be good to be clear -- ANOVA at the gene level, or are you > thinking of mixing in experiment-level parameters? For the former, > Biobase would suffice, though other packages (limma, nlme) might be > useful as well. This assumes that you've already got normalized data > (or else you'd need the appropriate packages) It's a two-way experiment. Young vs. old, mutant vs. wildtype, with 8 to 10 replicate arrays in each of the four possible combinations. Would that mean it's experiment-level or gene-level (sorry, I don't yet know the proper terminology)? The sort of thing that in SAS would involve PROC GLM... except the data is probably not normally distributed even after log2 (thus necessitating the empirical null distribution) and of course the multiple comparison problem (thus necessitating multtest or qvalue). I'm surprised that a two-way ANOVA using an empirical null dist and then adjusted p-values isn't the default analysis everyone does on their microarrays... could it be that everyone in the field is doing t-test style comparisons only? -------------------------------------- Protect yourself from spam, use http://sneakemail.com

"Alex F. Bokov" <yjih74b02@sneakemail.com> writes: > Wow, that was fast! Thanks for your answer. > > rossini-at-blindglobe.net (A.J. Rossini) |bioconductor.org| wrote: > >> It would be good to be clear -- ANOVA at the gene level, or are you >> thinking of mixing in experiment-level parameters? For the former, >> Biobase would suffice, though other packages (limma, nlme) might be >> useful as well. This assumes that you've already got normalized data >> (or else you'd need the appropriate packages) > > It's a two-way experiment. Young vs. old, mutant vs. wildtype, with > 8 to 10 replicate arrays in each of the four possible > combinations. Would that mean it's experiment-level or gene-level > (sorry, I don't yet know the proper terminology)? I'm assuming that you want to fit a 2-way ANOVA to each gene or tag. That would be easy, the overview being to stick the data into an exprSet, and then use esApply to compute p-values from which to evaluate via multtest. The question would be if you wanted to include a parameter for modeling/testing that was the same for all genes. > The sort of thing that in SAS would involve PROC GLM... except the > data is probably not normally distributed even after log2 (thus > necessitating the empirical null distribution) and of course the > multiple comparison problem (thus necessitating multtest or qvalue). > I'm surprised that a two-way ANOVA using an empirical null dist and > then adjusted p-values isn't the default analysis everyone does on > their microarrays... could it be that everyone in the field is doing > t-test style comparisons only? Depends on all of the following: the technology and chips used, the experiment (and different scientific areas seem to have different uses for expression arrays), and how well normalization works. Not all experiments are in the format you described (very simple 2-way layout), and not all observed datasets are appropriate for ANOVA. best, -tony -- A.J. Rossini / rossini@u.washington.edu / rossini@scharp.org Biomedical/Health Informatics and Biostatistics, University of Washington. Biostatistics, HVTN/SCHARP, Fred Hutchinson Cancer Research Center. FHCRC: 206-667-7025 (fax=4812)|Voicemail is pretty sketchy/use Email CONFIDENTIALITY NOTICE: This e-mail message and any attachments ... {{dropped}}