Hello,
I was trying to do a DESeq2 analysis on pyrosequence data that i have, and I'm trying different software in the DE analysis. In DEseq2 my DE genes were too low, my mean counts are low too since the median is less than 10, and i have 2 replicates in one condition and 4 in the other one. So I was trying to get a little more DE genes like the other softwares playing with the options and if i used the FitType=local then the DE went up a little bit, from another question answered here (https://support.bioconductor.org/p/81094/) I read that you could see if the dispersionplot got better or not. In my opinion it is better with local but I want to be sure and also be able to do it numerically, so how do I get the Genedispest? Is it something like this : mcols(dds)$dispGeneEst and ghow do i get the Dispfit, Sorry I'm kind of new with R and I would be thankful if somebody could help!
Here are my plots:
Local:https://ibb.co/TMsTSN1
Parametric: https://ibb.co/XD4qJc7
Thank you very much with the quick response! I'll use the normal pipeline
Thank you very much with the quick response! I'll use the normal pipeline
Hello Michael and sorry for the inconvenience, but as I was organizing my results and corroborating them I didn't pre-filter the data and use the normal pipeline, and in that normal pipeline the fitType automatically set itself to local, so if i filter or not pre-filter the data the functions used change, the pre-filtering I used was:
keep <- rowSums(counts(dds)>= 5) >= 4
I think its better to pre-filter but in one post(https://support.bioconductor.org/p/92455/) I read that it was unnecessary and that the packages used independent filtering internally, and it was mostly used to speed up the processing. So I don't know If I should be more cautious and examine Cooks distance better or just don't mind this observation
Pre-filtering is not a problem. You can pre-filter. But it’s hard to say what is the “right” choice per dataset. Mostly you don’t want to try too many choices post hoc in order to find small pvalues.