hi Jose, On Fri, Jan 17, 2014 at 11:52 AM, Garcia Manteiga Jose Manuel < garciamanteiga.josemanuel@hsr.it> wrote: > Dear Mike, > Thanks for the confirmation, I remember talking to someone during the > Bioc2013 lab saying that same thing on shrunken log2FC but I do not know > why I thought that the value to use with pre-ranked was not the log2FC in > the results object but another one due to the shrinking procedure. > > On the other hand, on the workflow from the lab, I remember that in 5.1 > log2FC was used to test whether particular gene sets had an average log2FC > different from 0. That would use log2FC information but in a different way > compared to how I understand GSEA pre-ranked analysis (the procedure > published by the people in Broad). In the Broad pre-ranked strategy, the > enrichment using log2FCs from the top (positive and negative) is giving you > an enrichment score that should be more related, in my view, to biological > relevance since it is linked to highest log2FCs observed, in absolute > values, whereas the workflow in 5.1 would flag those gene sets that do show > a different behaviour but maybe very low just because we are mathematically > able to say so (because of the t-test used). > > By the way, could the thresholded test implementation in DESeq2 >1.3, or > something similar, be used in a similar way to filter for those gene sets > with a more relevant “enrichment” using the 5.1 workflow? > Thanks again for the efforts > ​No, because using tests of log2 fold change with thresholds > 0 produces different p-values. The 5.1 workflow doesn't involve p-values at all.​ ​ Mike​ > > Jose > > > > On 17 Jan 2014, at 16:10, Michael Love <michaelisaiahlove@gmail.com> > wrote: > > hi Garcia, > > For DESeq2, we do recommend using shrunken log fold changes for gene set > tests, that is, the log2FoldChange column of a results object when the > default DESeq2 pipeline has been used. > > For a suggested workflow, see section 5.1 of the PDF for the presentation "DESeq2/DEXSeq > / Alejandro Reyes, Wolfgang Huber", one of the Bioc2013 labs: > http://bioconductor.org/help/course-materials/2013/BioC2013/ > > Mike > > > > On Fri, Jan 17, 2014 at 3:30 AM, Garcia Manteiga Jose Manuel < > garciamanteiga.josemanuel@hsr.it> wrote: > >> Hi, >> I wanted the same and this is what gsea people replied to me some months >> ago: >> >> >> Hi, >> >> Thank you for your interest in GSEA. >> >> R GSEA does accept ranked list as an input. >> >> Please note that R GSEA has not been actively maintained since 2005. >> >> Regards, >> >> >> I have not gone through the R GSEA again to check how pre-ranked lists >> can be accepted, but the last sentence in their message made me hold a >> minute and think in other possibilities. >> For the moment I am still doing as you mentioned, using the rnk file with >> java, then getting back again from results of gsea to R. >> >> By the way, I would like to known which metric should be best to be used >> for that kind of analysis when using RNA-Seq data coming from DESeq2 >> analysis. log2FC? p-values? Are they considered to be weighted, as GSEA >> pre-ranked names them? >> Thanks >> Regards >> >> >> >> On 17 Jan 2014, at 00:18, Daniel Schmolze <bioconductor@schmolze.com>> <mailto:bioconductor@schmolze.com>> wrote: >> >> I want to do a GSEA entirely from within R, using genes ranked by my >> own metric. At the moment I'm saving my ranked genes to a .rnk file, >> then calling Broad's GSEA java program, then reading the resulting >> output back into R (all I care about are the p-values in >> gsea_report_for_na_pos_####.xls and gsea_report_for_na_neg_####.xls). >> Cumbersome to say the least. >> >> As far as I can tell, the Broad GSEA R script won't accept pre- ranked >> genes, but maybe I'm wrong? If not, I'm interested in other options. >> I'd like to specifically stick with the Broad GSEA algorithm if >> possible. >> >> _______________________________________________ >> Bioconductor mailing list >> Bioconductor@r-project.org<mailto: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]] >> >> _______________________________________________ >> 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]]