which one is better for De analysis ? edgeR, DESeq2, voom or ROTS ?

Farbod Emami

unread,

Oct 21, 2015, 6:48:53 PM10/21/15

to trinityrnaseq-users

Hi friends,

I have used edgeR, DESeq2 and voom for "run_DE_analysis.pl" command for comapring the results and have created 3 directory with ".DE_results" files and "volcano" plots.

then I have run "analyze_diff_expr.pl" command with p-value 1e-3 of each one and with the similar Samples_described.txt file and similar Trinity_trans.TMM.EXPR.matrix file for each run (I repeat every thing was similar except the package I have used for run_DE_analysis)

why the DE transcripts is so different:

num-DE-features in voom : 75

num-DE-features in DESeq2 : 314

num-DE-features in edgeR : 36

which is robuster or more correct ? and why ?

Thanks

Farbod

Ken Field

unread,

Oct 21, 2015, 7:26:04 PM10/21/15

to Farbod Emami, trinityrnaseq-users

Farbod-

I am afraid that no one can answer that question for you. You will have to read the papers and decide for yourself:

http://www.biomedcentral.com/content/pdf/gb-2014-15-2-r29.pdf

http://www.genomebiology.com/2014/15/12/550

http://bioinformatics.oxfordjournals.org/content/26/1/139.short

Sincerely,

Ken

--
You received this message because you are subscribed to the Google Groups "trinityrnaseq-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to trinityrnaseq-u...@googlegroups.com.
To post to this group, send email to trinityrn...@googlegroups.com.
Visit this group at http://groups.google.com/group/trinityrnaseq-users.
For more options, visit https://groups.google.com/d/optout.

--

Ken Field, Ph.D.

Associate Professor of Biology

Program in Cell Biology/Biochemistry

Bucknell University

Room 203A Biology Building

Farbod Emami

unread,

Oct 22, 2015, 4:36:29 AM10/22/15

to trinityrnaseq-users, farbo...@gmail.com

Dear Ken, Hi and thank you

1- is it advisable that check the DE transcriptome between male and female RNA-seq with all of these 4 platforms and then select the uniqe ones and report it as a sum of truly DE transcriptome ? did you have a paper with this strategy ? (I mean in this case - -> 74+314+36 - (repeated ones) = the most reliable DEG )

2- As I have realized most of the de novo RNA-seq papers count on the DE transcripts, but you know that the Trinity package produce a DE profiles of genes, too (which in my case the number of them is bigger that the DE transcripts), why there is not any attention on DEG in de novo transcriptome analysis paper compare to DE transcripts?

Thank you again

On Thursday, October 22, 2015 at 2:56:04 AM UTC+3:30, Ken Field wrote:

Farbod-
I am afraid that no one can answer that question for you. You will have to read the papers and decide for yourself:
http://www.biomedcentral.com/content/pdf/gb-2014-15-2-r29.pdf
http://www.genomebiology.com/2014/15/12/550
http://bioinformatics.oxfordjournals.org/content/26/1/139.short

Sincerely,
Ken

On Wed, Oct 21, 2015 at 6:48 PM, Farbod Emami <farbo...@gmail.com> wrote:

Hi friends,
I have used edgeR, DESeq2 and voom for "run_DE_analysis.pl" command for comapring the results and have created 3 directory with ".DE_results" files and "volcano" plots.
then I have run "analyze_diff_expr.pl" command with p-value 1e-3 of each one and with the similar Samples_described.txt file and similar Trinity_trans.TMM.EXPR.matrix file for each run (I repeat every thing was similar except the package I have used for run_DE_analysis)

why the DE transcripts is so different:

num-DE-features in voom : 75
num-DE-features in DESeq2 : 314
num-DE-features in edgeR : 36

which is robuster or more correct ? and why ?
Thanks
Farbod

--
You received this message because you are subscribed to the Google Groups "trinityrnaseq-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to trinityrnaseq-users+unsub...@googlegroups.com.

To post to this group, send email to trinityrn...@googlegroups.com.
Visit this group at http://groups.google.com/group/trinityrnaseq-users.
For more options, visit https://groups.google.com/d/optout.

Ken Field

unread,

Oct 22, 2015, 7:48:20 AM10/22/15

to Farbod Emami, trinityrnaseq-users

Dear Farbod-

I have seen some papers that take that approach, or to report all results along with a venn diagram and focus on only those results that are in common for all methods.

However, I don't like either of those because each method makes different statistical assumptions about the data and handles outliers and biases in your data differently. If the data were ideal (i.e. from a simulation) then all three should give very similar answers and these approaches would be valid. However, with real data it is common that some of the assumptions of a model will not apply perfectly. For that reason, you will get different answers from each of the methods and some of the methods will contain more type I errors while others will contain more type II errors. Unfortunately, you can't tell just by looking at your lists of genes which are which.

For these reasons, what I did, was to explore the results of multiple methods, like you suggest, but then determine which method produced results that I was most confident in.

http://journals.plos.org/plospathogens/article?id=10.1371/journal.ppat.1005168

In the paper only those results are reported. I didn't use voom, but I compared DESeq2, edgeR, and EBSeq. At the "gene" level, I found that DESeq2 produced a very nice (symmetrical) MA plot and a list of genes that I could be very confident in (after exploring the data carefully). Unfortunately, using DESeq2, I lost a few interesting genes that were removed due to the outlier procedure, but it was hard to be confident in those genes anyway. I also used EBSeq to look at transcript-level DE. I was more comfortable with the Bayesian approach to isoform uncertainty taken by that method and the results complemented those of DESeq2 nicely. However, I think I would have found almost exactly the same results if I had used DESeq2 for both.

From what you described about your results, I would say that things are looking good for DESeq2. That is considered by its authors to be a very conservative algorithm and they recommend starting at a FDR cutoff of 0.1. I have seen papers that use 0.1 or 0.05 (what I used in the PLOS Pathogens paper) or more stringent cutoffs. You should run your data through the entire DESeq2 vignette to see how it looks compared to the sample data. This will help you decide whether it is meeting the assumptions of that statistical model. If it looks good after that, then you can have confidence in selecting DESeq2 for analysis. I would run it at different FDR cutoffs to see how that affects the genes that you have and choose a cutoff that you are comfortable with based on the numbers of genes and the expected false discoveries.

I hope that this helps.

Sincerely,

Ken

To unsubscribe from this group and stop receiving emails from it, send an email to trinityrnaseq-u...@googlegroups.com.

To post to this group, send email to trinityrn...@googlegroups.com.
Visit this group at http://groups.google.com/group/trinityrnaseq-users.
For more options, visit https://groups.google.com/d/optout.

--
Ken Field, Ph.D.
Associate Professor of Biology
Program in Cell Biology/Biochemistry
Bucknell University
Room 203A Biology Building

--

You received this message because you are subscribed to the Google Groups "trinityrnaseq-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to trinityrnaseq-u...@googlegroups.com.

To post to this group, send email to trinityrn...@googlegroups.com.
Visit this group at http://groups.google.com/group/trinityrnaseq-users.
For more options, visit https://groups.google.com/d/optout.

Farbod Emami

unread,

Oct 22, 2015, 8:16:10 AM10/22/15

to trinityrnaseq-users, farbo...@gmail.com

Dear Dr. Ken

Thank you, it was very useful and informative for me.

an a honor for me speaking with the author of such magnificent paper!

Thanks a lot

To unsubscribe from this group and stop receiving emails from it, send an email to trinityrnaseq-users+unsub...@googlegroups.com.

To post to this group, send email to trinityrn...@googlegroups.com.
Visit this group at http://groups.google.com/group/trinityrnaseq-users.
For more options, visit https://groups.google.com/d/optout.

--
Ken Field, Ph.D.
Associate Professor of Biology
Program in Cell Biology/Biochemistry
Bucknell University
Room 203A Biology Building

--
You received this message because you are subscribed to the Google Groups "trinityrnaseq-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to trinityrnaseq-users+unsub...@googlegroups.com.

To post to this group, send email to trinityrn...@googlegroups.com.
Visit this group at http://groups.google.com/group/trinityrnaseq-users.
For more options, visit https://groups.google.com/d/optout.

Reply all

Reply to author

Forward