Dear all, I am doing differential expression analysis and I have a question concerning time course experiments (Single-Channel Experimental Designs). I have one cell line that was treated in 4 different ways. I want to check which genes respond di erently over time for different treatments. I did 4 different comparisons. I have treatment A, B, C and D, and I compared groups: A-B, A-C, C-D and B-D. For all my data I created ONE design matrix, and FOUR contrast.diff.matrices. For the fit() function I have used the esetPROC with all my data. This was followed by contrast.fit() and eBayes() functions. At the end I got top differentially expressed genes (from topTableF() function). Additionally, I did almost the same thing, but I created FOUR different design matrices and FOUR contrast.diff.matrices for all my comparisons. I extracted the subset of esetPROC only with the data I needed for the comparison, and continued as described above. I got different results for those two approaches. The adj.p.values were much smaller for the first approach than for the second one. I assume it is because of the eBayes function. Could you please explain me which approach is the correct/better one and why? Best wishes, Agata -- output of sessionInfo(): R version 3.0.2 (2013-09-25) Platform: x86_64-unknown-linux-gnu (64-bit) locale: [1] C attached base packages: [1] parallel stats graphics grDevices utils datasets methods [8] base other attached packages: [1] gplots_2.12.1 lattice_0.20-24 sva_3.8.0 [4] mgcv_1.7-26 nlme_3.1-111 corpcor_1.6.6 [7] vsn_3.30.0 marray_1.40.0 hgug4112a.db_2.10.1 [10] org.Hs.eg.db_2.10.1 Agi4x44PreProcess_1.22.0 genefilter_1.44.0 [13] annotate_1.40.0 AgiMicroRna_2.12.0 affycoretools_1.34.0 [16] KEGG.db_2.10.1 GO.db_2.10.1 RSQLite_0.11.4 [19] DBI_0.2-7 AnnotationDbi_1.24.0 preprocessCore_1.24.0 [22] affy_1.40.0 Biobase_2.22.0 BiocGenerics_0.8.0 [25] biomaRt_2.18.0 limma_3.18.12 WriteXLS_3.4.0 loaded via a namespace (and not attached): [1] AnnotationForge_1.4.4 BSgenome_1.30.0 BiocInstaller_1.12.0 [4] Biostrings_2.30.1 Category_2.28.0 DESeq2_1.2.10 [7] Formula_1.1-1 GOstats_2.28.0 GSEABase_1.24.0 [10] GenomicFeatures_1.14.2 GenomicRanges_1.14.4 Hmisc_3.14-0 [13] IRanges_1.20.6 KernSmooth_2.23-10 MASS_7.3-29 [16] Matrix_1.1-2 PFAM.db_2.10.1 R.methodsS3_1.6.1 [19] R.oo_1.17.0 R.utils_1.29.8 R2HTML_2.2.1 [22] RBGL_1.38.0 RColorBrewer_1.0-5 RCurl_1.95-4.1 [25] Rcpp_0.11.0 RcppArmadillo_0.4.000.2 ReportingTools_2.2.0 [28] Rsamtools_1.14.3 VariantAnnotation_1.8.12 XML_3.98-1.1 [31] XVector_0.2.0 affyio_1.30.0 annaffy_1.34.0 [34] biovizBase_1.10.7 bit_1.1-11 bitops_1.0-6 [37] caTools_1.16 cluster_1.14.4 codetools_0.2-8 [40] colorspace_1.2-4 dichromat_2.0-0 digest_0.6.4 [43] edgeR_3.4.2 ff_2.2-12 foreach_1.4.1 [46] gcrma_2.34.0 gdata_2.13.2 ggbio_1.10.11 [49] ggplot2_0.9.3.1 graph_1.40.1 grid_3.0.2 [52] gridExtra_0.9.1 gtable_0.1.2 gtools_3.3.0 [55] hwriter_1.3 iterators_1.0.6 labeling_0.2 [58] latticeExtra_0.6-26 locfit_1.5-9.1 munsell_0.4.2 [61] oligoClasses_1.24.0 plyr_1.8 proto_0.3-10 [64] reshape2_1.2.2 rtracklayer_1.22.3 scales_0.2.3 [67] splines_3.0.2 stats4_3.0.2 stringr_0.6.2 [70] survival_2.37-7 tools_3.0.2 xtable_1.7-1 [73] zlibbioc_1.8.0 packageDescription('limma')$Maintainer [1] "Gordon Smyth <smyth at="" wehi.edu.au="">" -- Sent via the guest posting facility at bioconductor.org.

Hi Agata, On 3/28/2014 5:18 AM, Agata [guest] wrote: > Dear all, > > I am doing differential expression analysis and I have a question concerning time course experiments (Single-Channel Experimental Designs). > > I have one cell line that was treated in 4 different ways. I want to check which genes respond di erently over time for different treatments. I did 4 different comparisons. > > I have treatment A, B, C and D, and I compared groups: A-B, A-C, C-D and B-D. For all my data I created ONE design matrix, and FOUR contrast.diff.matrices. For the fit() function I have used the esetPROC with all my data. This was followed by contrast.fit() and eBayes() functions. At the end I got top differentially expressed genes (from topTableF() function). > > Additionally, I did almost the same thing, but I created FOUR different design matrices and FOUR contrast.diff.matrices for all my comparisons. I extracted the subset of esetPROC only with the data I needed for the comparison, and continued as described above. > > I got different results for those two approaches. The adj.p.values were much smaller for the first approach than for the second one. I assume it is because of the eBayes function. Could you please explain me which approach is the correct/better one and why? It's a combination of the eBayes function and the fact that you estimate gene-wise variances using all the data, rather than a subset (in the first analysis, that is). Both methods are technically correct, but as you have already noted, the first method is more powerful, so in my mind it is to be preferred. Best, Jim > > Best wishes, > Agata > > > > -- output of sessionInfo(): > > R version 3.0.2 (2013-09-25) > Platform: x86_64-unknown-linux-gnu (64-bit) > > locale: > [1] C > > attached base packages: > [1] parallel stats graphics grDevices utils datasets methods > [8] base > > other attached packages: > [1] gplots_2.12.1 lattice_0.20-24 sva_3.8.0 > [4] mgcv_1.7-26 nlme_3.1-111 corpcor_1.6.6 > [7] vsn_3.30.0 marray_1.40.0 hgug4112a.db_2.10.1 > [10] org.Hs.eg.db_2.10.1 Agi4x44PreProcess_1.22.0 genefilter_1.44.0 > [13] annotate_1.40.0 AgiMicroRna_2.12.0 affycoretools_1.34.0 > [16] KEGG.db_2.10.1 GO.db_2.10.1 RSQLite_0.11.4 > [19] DBI_0.2-7 AnnotationDbi_1.24.0 preprocessCore_1.24.0 > [22] affy_1.40.0 Biobase_2.22.0 BiocGenerics_0.8.0 > [25] biomaRt_2.18.0 limma_3.18.12 WriteXLS_3.4.0 > > loaded via a namespace (and not attached): > [1] AnnotationForge_1.4.4 BSgenome_1.30.0 BiocInstaller_1.12.0 > [4] Biostrings_2.30.1 Category_2.28.0 DESeq2_1.2.10 > [7] Formula_1.1-1 GOstats_2.28.0 GSEABase_1.24.0 > [10] GenomicFeatures_1.14.2 GenomicRanges_1.14.4 Hmisc_3.14-0 > [13] IRanges_1.20.6 KernSmooth_2.23-10 MASS_7.3-29 > [16] Matrix_1.1-2 PFAM.db_2.10.1 R.methodsS3_1.6.1 > [19] R.oo_1.17.0 R.utils_1.29.8 R2HTML_2.2.1 > [22] RBGL_1.38.0 RColorBrewer_1.0-5 RCurl_1.95-4.1 > [25] Rcpp_0.11.0 RcppArmadillo_0.4.000.2 ReportingTools_2.2.0 > [28] Rsamtools_1.14.3 VariantAnnotation_1.8.12 XML_3.98-1.1 > [31] XVector_0.2.0 affyio_1.30.0 annaffy_1.34.0 > [34] biovizBase_1.10.7 bit_1.1-11 bitops_1.0-6 > [37] caTools_1.16 cluster_1.14.4 codetools_0.2-8 > [40] colorspace_1.2-4 dichromat_2.0-0 digest_0.6.4 > [43] edgeR_3.4.2 ff_2.2-12 foreach_1.4.1 > [46] gcrma_2.34.0 gdata_2.13.2 ggbio_1.10.11 > [49] ggplot2_0.9.3.1 graph_1.40.1 grid_3.0.2 > [52] gridExtra_0.9.1 gtable_0.1.2 gtools_3.3.0 > [55] hwriter_1.3 iterators_1.0.6 labeling_0.2 > [58] latticeExtra_0.6-26 locfit_1.5-9.1 munsell_0.4.2 > [61] oligoClasses_1.24.0 plyr_1.8 proto_0.3-10 > [64] reshape2_1.2.2 rtracklayer_1.22.3 scales_0.2.3 > [67] splines_3.0.2 stats4_3.0.2 stringr_0.6.2 > [70] survival_2.37-7 tools_3.0.2 xtable_1.7-1 > [73] zlibbioc_1.8.0 > > > packageDescription('limma')$Maintainer > [1] "Gordon Smyth <smyth at="" wehi.edu.au="">" > > -- > Sent via the guest posting facility at bioconductor.org. > > _______________________________________________ > 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