Hi, I used dexseq_count.py to generate a count set. I used read.HTSeqCounts to get an ExonCountSet I manage to go throught the vignette just fine, but I noticed that a concatenated gene (ENSG+ENSG+ENSG) with 84 exons is not testable. I have data for 6 controls and 4 patients. If I sum up their counts, the median over all exons is about 2000 and the mean about 5000. This should be way enough to make a gene testable. However, it was not. This is in the vignette: Before starting estimating the CR dispersion estimates,estimateDispersionsfi rst defi nes the \ testable" counting bins. In this step, all bins are excluded with less that minCount reads. By default, minCount=10, i.e., the bin must have at least 10 reads, summed up across all samples, but other values can be specifi ed when calling estimateDispersion. If a gene has only one testable counting bin, then this counting bin is then considered as not testable, either, because it's usage cannot be compared to any other counting bins. The testable bins are marked in the column testable of the feature data. With 82 testing bins, this gene should have turned out to be testable. I removed the two exons with 1 and 3 reads in total, and rerun the program. This time the gene turns up to be testable. Now, I was highly interested in this particular gene, but I now wonder whether I have to remove all exons < 10 reads, in order for the program to take along all genes that still might be interesting to look at. Or should I just put minCount to 1? Hanneke [[alternative HTML version deleted]]

Dear Hanneke On 04/11/13 11:55, Hanneke van Deutekom wrote: > With 82 testing bins, this gene should have turned out to be testable. I > removed the two exons with 1 and 3 reads in total, and rerun the program. > This time the gene turns up to be testable. > Now, I was highly interested in this particular gene, but I now wonder > whether I have to remove all exons < 10 reads, in order for the program to > take along all genes that still might be interesting to look at. Or should > I just put minCount to 1? No, you are right, of course, that DEXSeq should have marked only the two low-count exons as not testable and keep all the others in the test. Maybe you stumbled over a strange bug. Could you perhaps send us by email your ExconCountSet, so we can investigate? Please let us also know which gene you mean and which DEXSeq version you are using. Simon