hi Jianming, Let's keep the discussion on the bioconductor list. On Wed, Sep 4, 2013 at 10:43 AM, 邵建明 <jianmingshao1987@gmail.com> wrote: > Hi Mike, > Thank you for your mail about my question. The 83 genes were > selected based on the GWAS results, so I did not know genes' expression > pattern between cases and controls, and that was what I want to know from > RNA-capture sequencing of 83 genes. The sequencing depth could be > normalized by RPKM, the traditional RNA-seq gene expression normalization > method which normalize gene expression by dividing gene length and total > reads number. > ​You can go ahead with a differential expression analysis, but keep in mind the following problem, given that you selected a small set of genes which are candidates for differential expression. say you have the following counts, for 2 control, 2 case samples: gene_A: 2 2 2 2 gene_B​: 2 2 1 1 What you cannot tell apart is whether the size factors should be 2,2,1,1 (in which case gene A has a fold change of 2) or should be 2,2,2,2 (in which case gene B has a fold change of 1/2). In typical RNA-Seq experiments, all genes are assayed, so you can more reliably estimate size factors by assuming that some subset of genes do not change expression level (by using medians or trimmed means) . See for instance the section on estimate size factors with median ratios here http://genomebiology.com/2010/11/10/r106 or the TMM normalization method here http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2864565/. If you were to perform a differential analysis with DESeq and then examine the results with plotMA(), you wouldn't know exactly where the 0 on the y axis should go. One solution to this would be using spike-in controls. However, given the dataset you have, hopefully most of the log2 fold changes would be small (e.g. less than 0.1). If the log2 fold changes were spread out farther it would be difficult to draw definitive conclusions. Mike > 2013/9/4 Michael Love <michaelisaiahlove@gmail.com> > >> hi Jianming, >> >> How many of these 83 candidate genes do you expect to be differentially >> expressed? >> >> The problem is that: imagine that in the case samples, all 83 candidate >> genes are upregulated. Without some way of assessing the sequencing depth >> (i.e. normalization using spike-in controls) it would be impossible to tell >> apart differential expression from sequencing depth. >> >> Mike >> >> >> On Tue, Sep 3, 2013 at 10:35 AM, 邵建明 <jianmingshao1987@gmail.com> wrote: >> >>> Dear all, >>> I am a PhD candidate in Beijing Institute of Genomics, Chinese >>> Academy of Sciences. Recently I have been worked with data analysis >>> concerning RNA capture followed by high throughput sequencing. Four >>> samples, 2 cases and 2 controls, were used for sequencing. My library >>> preparation protocol is similar to workflow of Exome capture, except for >>> the material used for capture was cDNA and the capture library was >>> customized probes synthesized by Agilent. After mapping, I want to do >>> DEG >>> analysis utilizing DESeq, and I found the gene number would affect the >>> results given by DESeq. So my question is whether DESeq compatible with >>> limited genes (83 candidate genes for my project)? And would you please >>> give me some suggestions about DEG analysis concerning candidate genes' >>> RNA-seq? or, for my project, I could just calculate RPKM value for each >>> gene, and identify DEGs simply by fold change > 2? Thank you! >>> Sincerely, >>> >>> Jianming SHAO >>> >>> [[alternative HTML version deleted]] >>> >>> _______________________________________________ >>> Bioconductor mailing list >>> Bioconductor@r-project.org >>> https://stat.ethz.ch/mailman/listinfo/bioconductor >>> Search the archives: >>> http://news.gmane.org/gmane.science.biology.informatics.conductor >>> >> >> > [[alternative HTML version deleted]]