Lana Schaffer wrote: > Hi, > This is just a case of matching profiles during something > like a time course. In packages like GeneSpring and Spotfile the user > is allowed to chose a profile and then find other "genes" with matching > profiles for varying correlations. In this case I have pvalues that are > significant for differential expression at short time periods and not > significant at long time periods. I want to get the list of genes with that > profile. genefinder in the genefilter package - does something like this. > Alternatively, the log expression ratio is high at short time period and > then levels off > at long time periods for all the genes of interest. I thought that the high > pvalues > was due to increased variance and therefore heterogeneity. I don't know how > to > think about the decreased expression along with this, since I am dealing > with > differential expression. I have done co-expression analysis for all these > genes and > find them to be co-expressed between 2 modules. Show the goal is to show > levels of hetergeneity between the time periods. > > I am wondering if I used limma correctly for I divided up the samples into > 3 "time periods" and then then fit the samples together. I then used > contrasts to get adjusted pvalues for the genes for the 3 "time periods". > When > I graphed the trends in pvalues for each of the genes over time I get > profiles which > increase and then flatten for a set of genes (I want to get that set of > genes) > and then other profiles. I want to show that the > variance (hetergeneity) increases with time with some of the genes. I do not understand your pre-occupation with p-values. I think you should be interested in patterns of expression, not patterns in the p-values. > I think that I could do a multivariate regression to indicate a regression > in > differential expression, but then if the ratio is leveling off then > regression > won't tell me anything. That is why you need to fully specify the profile of interest, and then measure distances from it. > I hope you can understand where I am going. > Lana > > > ----- Original Message ----- > From: "Robert Gentleman" <rgentlem at="" fhcrc.org=""> > To: "Lana Schaffer" <schaffer at="" scripps.edu=""> > Cc: <bioconductor at="" stat.math.ethz.ch=""> > Sent: Tuesday, September 26, 2006 9:02 AM > Subject: Re: [BioC] batch effect on variances > > >> >> Lana Schaffer wrote: >>> Hi, >>> I want to find out if there is a batch effect (FEM or REM) on the >>> variance for 2 sets of >>> data which are discrete (different) treatments (time). The GeneMeta >>> package is designed >>> to combine batches which measure the same treatment effects. However, I >>> have what >>> corresponds to 2 different treatment effects. Is it valid to check >>> homogeneity for the 2 batches? >> Hi, >> You can do some things, but I am not sure why you care? If the two >> experiments do not have the same treatments then there is no sensible >> analysis that combines them, so whether or not the variances are the same, >> seems like an odd question, at least to me. What would you want to say >> about it and how might you try and use it? >> >> You can fit an appropriate model to each gene in each experiment >> separately, say using limma or any of the multitude of packages in BioC to >> do this. Once that has been done, you can estimate per gene variances, and >> then their ratio, suitably normalized will almost surely follow some form >> of F statistic (provided that samples are not too small and that the >> models are reasonable). But I am still not sure what you would do with >> such information. >> >> best wishes >> Robert >> >> >>> Lana Schaffer >>> Biostatistics/Informatics >>> The Scripps Research Institute >>> DNA Array Core Facility >>> La Jolla, CA 92037 >>> (858) 784-2263 >>> (858) 784-2994 >>> schaffer at scripps.edu >>> >>> >>> >>> [[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 >>> >> -- >> Robert Gentleman, PhD >> Program in Computational Biology >> Division of Public Health Sciences >> Fred Hutchinson Cancer Research Center >> 1100 Fairview Ave. N, M2-B876 >> PO Box 19024 >> Seattle, Washington 98109-1024 >> 206-667-7700 >> rgentlem at fhcrc.org >> > > _______________________________________________ > 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 > -- Robert Gentleman, PhD Program in Computational Biology Division of Public Health Sciences Fred Hutchinson Cancer Research Center 1100 Fairview Ave. N, M2-B876 PO Box 19024 Seattle, Washington 98109-1024 206-667-7700 rgentlem at fhcrc.org

On Tuesday 26 September 2006 13:29, Lana Schaffer wrote: > Hi, > This is just a case of matching profiles during something > like a time course. In packages like GeneSpring and Spotfile the user > is allowed to chose a profile and then find other "genes" with matching > profiles for varying correlations. In this case I have pvalues that are > significant for differential expression at short time periods and not > significant at long time periods. I want to get the list of genes with > that profile. Hi, Lana. You might want to choose your genes not based on the p-value for a particular time point, but rather based on an f-statistic, of which there is only one per gene. Then, do whatever type of clustering you like using those genes that show a high f-statistic. I think a lot of folks would use kmeans or hierarchical clustering to then group genes that show a similar profile, but there are many ways to approach clustering. Sean