cufflinks 2.0.0

[MeixiaZhao]

I have just used the latest cufflinks 2.0.0 to run my previous data, as I want to detect if large differences occur. And I did find some very confuse results. For the significant differential expressed isoforms, the number of cufflinks 1.3.0 is 11380, but the number from cufflinks 2.0.0 reduced to 2371. And for the significant differential expressed genes is the same situation, 2692 vs 1353, respectively. Many genes or transcripts showed significant change in previous version, but now changed from "yes" to "no". As my data has no replicates, first I thought this may due to the option --min-reps-for-js-test, but even I changed the default 3 to 1, the results is same. I don't know if for publication, the results from which version I can use due to the large difference? Did anyone meet this kind of situation?
Thanks!

[sdriscoll]

by design, cufflinks 2.0 will produce more conservative gene lists than previous versions. I think previous versions had a very high false discovery rate. i've only used cuffdiff from cufflinks 2.0 on two different sets of data but so far I'd say it's maybe overly strict..but the results make more sense than they did in v1.3.

keep in mind that p-values on a N=1 comparison are kind of like magic. there's obviously very little statistical power there and DE tools should probably be overly conservative in that case. your data must have quite a bit of differential expression to produce a list that long with no replicates.

the --min-reps-for-js-test option only applies to the splicing.diff, cds.diff and promoters.diff files. they found that with less than 3 replicates the false discovery rate in those files was excessively high.

[MeixiaZhao]

Yeah, I understand. I'm just in a dilemma as the data showed so large differences.

[sdriscoll]

If you've made decisions already based on the old results just stick to those. I think the new list is probably safer and for sure if the first list contained false positives they have been dropped from the new list. If it's not too much trouble for you I'd use the new list...unless of course it removed some genes of interest! That brings up another point though...right? Those p-values don't tell us the biological significance.

[luigra]

Hi all,

I tested cufflinks 2.0.0 and I quickly found that it has some problems (certainly more than the previous version): whit the same Data Set In the version 1.3.0 I obtained in the isoform_exp.diff file roughly the 40% of FAIL TEST. In the current version there are no FAIL but the number of NOTEST increased enormously (roughly to 90%) with the consequent reduction of significant DE isoforms. I would not believe to the results of cuffdiff 2.0.0