[Simon Anders]

Sorry Simon, would you mind commenting on the new Cufflinks and Cuffdiff? I don't mean to put you on the spot, since there is clearly a competition aspect of this, but in this nascent field, there are so few experts that can comment authoritatively on this.

Well, for our take on it, see the Discussion section in our DEXSeq paper (Anders, Reyes, Huber; Genome Res., 2012).

Annoyingly, Genome Research still has not managed to put our supplement online, which might be relevant; hence, please look at it in the in the PDF available here. (Same PDF as the preprint, at the end.)

[turnersd]

Well, for our take on it, see the Discussion section in our DEXSeq paper (Anders, Reyes, Huber; Genome Res., 2012).

Annoyingly, Genome Research still has not managed to put our supplement online, which might be relevant; hence, please look at it in the in the PDF available here. (Same PDF as the preprint, at the end.)

Thanks Simon.

Based on many threads here regarding huge differences folks are seeing in the new version of cufflinks/cuffdiff, it would be nice to see the comparisons in tables S1 and S2 supplemented with results from Cufflinks 2.0.2.

[billstevens]

Thank you Simon.

To echo turnersd, it would be very, very nice to have a comparison to the new Cufflinks/Cuffdiff, but the new version did JUST come out. Have you had a chance to take a peak at it at all, Simon?

This article was very useful in that I can now safely ignore the results that I got with the old version of cufflinks. Cole has stated the new version is much, much more accurate. However, I just used Cufflinks and Cuffdiff 2.0.2, and I have found the results more than little confusing in regards to FPKM and values (and yes, I'm aware of the common scale transformation). I'm going to give the Tophat results to EasyRNASeq and then run it on DESeq, and then compare what I've got.

[areyes]

Dear billstevens and turnersd,

Actually, what you guys mention about the new versions of cuffdiff was also asked by the reviewers, and that is the reason we included the most recent versions of cuffdiff (by the time we did the resubmission) in the comparisons with DEXSeq. Also, please note that all the data and code used for the DEXSeq paper (for now only in the preprint) is available online in the Supplement II (http://www-huber.embl.de/pub/DEXSeq/...ent_II_v2.html), and because is hard to track the new versions of cuffdiff, any contribution of comparisons (with this or other data) would be greatly appreciated!

However, I just made the comparison using the latest version of cuffdiff (2.0.2) and this are the results of some of the comparisons (turnersd, this would be the continuation of the supplement table of the paper):

comparison                   cuffdiff_202_sp       DEXSeq_1.0.5
all-treated-vs-untreated        21                          159
untreated13-vs-untreated24      0                            8
untreated14-vs-untreated23      0                            7

The new cuffdiff seems to be very, very conservative.

[dpryan]

Out of curiosity, are the 21 findings from the all-treated-vs-untreated cuffdiff output also found in the DEXSeq results?

[areyes]

5 of them are not!

Alejandro

ps. I just noted that I got a "thumbs down" in my last post", probably because extensive editing ? :P

[billstevens]

Thanks so much areyes!

I'm just checking out my data with EasyRNASeq and DESeq, and I'll post a comparison between that and Cuffdiff 2.0.2 and Cuffdiff 1.3.

[kenietz]

Hi guys,
this thread was really informative to me. I have one question tho. I have PE rna-seq from a species without ref genome and annotation. I have only 3 samples coming from different conditions and no replicates.

Is it possible to use DESeq for DE analysis?

Thank you!

[dvanic]

You have no reference genome or annotation, and no replicates? That is NOT a good place to be in, and I would strongly advise you to "get some replicates" if you can.

However, the first problem I am assuming you have is doing de novo transcriptome assembly, for which I have had good experience with Trinity. The Trinity website also has a guide for how to do DE analysis in R (http://trinityrnaseq.sourceforge.net..._analysis.html). So, yes, you can do it, but with one replicate your tests would be very underpowered.

Hope this helps.

[kenietz]

Hi dvanic,
thank you for the info.

Well in fact the 3 samples were sequenced 2 times because of some trouble with the machine. The first run seems to be better tho. So i can say that i got replicates

My samples are from fish but when i used zebra fish as ref genome for tophat, cufflinks, cuffdiff it seems that my species is not very close to zebra fish. There was another genome a bit closer but still not perfect case. So yeah i did denovo transcriptome assembly but with soapdenovo-trans.

Now whats bugging me is the following. how can one do DE on 3 sets of transcriptome data when the contigs from the different sets have different IDs? I mean every assembly is creating some contigs and there are named with some IDs which will differ between the assemblies.

[sudders]

Hi dvanic,
Now whats bugging me is the following. how can one do DE on 3 sets of transcriptome data when the contigs from the different sets have different IDs? I mean every assembly is creating some contigs and there are named with some IDs which will differ between the assemblies.

As far as I can see you have two alternatives here. First you could use some kind of alignment to match the contigs across samples. Maybe best reciprical blast hits? A second alternative would be pool all three samples into one and assemble them all together to generate one assembly. Then quantify the transcripts in the joint assembly using the indevidual samples.

BTW, it sounds like you replicates are technical rather than biological replicates. It is generally not a good idea use technical replicates in DE analysis, as the models are explicitly designed to measure biological variance (which has a different distribution to technical variance). If you want to use your other run, I see no harm in pooling the two runs for each sample together.

[kenietz]

I see. The problem is that i cant pool them together and assemble. Too large sets. One of them is like 75M 2x75 PE and even soap-denovotrans seg faulted after loading more than 100M reads. So i had to remove some of the reads in order to assemble at all.The other sets are a bit smaller but still above 35M 75x2 reads each.

Seems that the first idea will do the job hopefully. It will take some time tho as the biggest contig set has like 40K contigs which are above 500bp. Well thats life i suppose

[naman]

RDA Analysis

I have few queries regarding the RDA analysis to discuss. Although I see clear difference in the bacterial community between the samples, I am not able to relate the OTU abundance table with bacterial composition data and my factors (Mutations/Day effect in my case) on the RDA axis.

It would be great if it can be explained on mothur shared file.

Cheers,

[sdriscoll]

I have few queries regarding the RDA analysis to discuss. Although I see clear difference in the bacterial community between the samples, I am not able to relate the OTU abundance table with bacterial composition data and my factors (Mutations/Day effect in my case) on the RDA axis.

It would be great if it can be explained on mothur shared file.

Cheers,

i don't have an answer but may I suggest starting a new thread? this one's about Cuffdiff vs DESeq...most people aren't going to see your question unless they are first interested in Cuffdiff vs DESeq...

[colaneri]

Using the genes.count-tracking file from cuffdiff in DESeq

DESeq needs count data in the form of rectangular table. My question is whether is correct or possible to use the genes.count_tracking file generated with cuffdiff as the counts table that DESeq requires?

I will appreciate your help
Alejandro