Bioconductor edger manual

Thank you for your attention, Tiago Jesus -- output of sessionInfo : none -- Sent via the guest posting facility at bioconductor. Hi Tiago, Just a quick comment? You do not mention whether you have replicates, but if you do, then it is generally better to use them to estimate the dispersion. If you do not have replicates, then you have to use your judgement. In addition, take the P-values with a grain of salt as it says in the manual, "Note that the p-values obtained and the number of signi cant genes will be very sensitive to the dispersion value chosen".

Best, Mark On Here we see that a single estimate for the coefficient of variation is a bad model since tagwise dispersion does not follow the model but instead increases as the counts per million CPM increases. Fitting a model in edgeR takes several steps. First, you must fit the common dispersion. Then you need to fit a trended model if you do not fit a trend, the default is to use the common dispersion as a trend.

Then you can fit the tagwise dispersion which is a function of this model. In addition to the common and tagwise disperson, we can also estimate a generalized linear model glm fit using edgeR.

In the same way that we've been doing above, we will just add these as additional data to the object we've been working with. Note that the tagwise biological coefficient of variation is different in this case than it was when we just estimated the common dispersion in the naive method above.

This is because we model the tagwise dispersions based on the model derived from the glm model that we choose. If we change the method power above to something else, the tagwise errors change to reflect that the method is different.

We have tags, so the default is bin. When I used the bin. DESeq always only uses a gamma glm as its model. Since edgeR does not have gamma glm as an option, we cannot produce the same glm results in edgeR as we can in DESeq and vice versa.

Note that this plots dispersion on the vertical axis instead of the biological coefficient of variation. The dispersion of a gene is simply another measure of a gene's variance and it is used by DESeq to model the overall variance of a gene's count values. The dispersion can be interpreted as the square of the coefficient of biological variation e.

Once the dispersions are estimated, we can proceed with testing procedures for determining differential expression. Documentation » Bioconductor Package vignettes and manuals. Workflows for learning and use. Course and conference material. Community resources and tutorials.

Support » Please read the posting guide. Post questions about Bioconductor to one of the following locations: Support site - for questions about Bioconductor packages Bioc-devel mailing list - for package developers.