Hi, Thanh,

If your primary goal is inference about differential 'gene' expression taking biological variability into account with biological replicates for each of two conditions, you might want (eg see Dillies et al., http://bib.oxfordjournals.org/content/early/2012/09/15/bib.bbs046.long and http://wiki.galaxyproject.org/Events/GCC2013/Abstracts#Events.2FGCC2013.2FAbstracts.2FPosters.P4:_Comparing_R-based_methods_and_Cuffdiff2_for_analysis_of_RNA-seq_data_in_Galaxy) to try (and compare!) edgeR (and optionally DESeq and VOOM/limma). A set of *very much beta* tools is available for admin installation and user testing from the test toolshed in the statistics section owned by fubar. 

The edgeR tool can optionally run 2 way GLM. It requires raw count matrices as inputs which can be generated from a GTF/'gene' model of your choice and any number of mapped SAM/BAM inputs using the htseq based companion tool in the same tool shed section. Please don't install to a production machine yet but we're getting good results from it - feedback and code improvements are welcomed from willing beta testers. 

The R 3.0.x tool shed dependency package in particular is still under development and is likely to change substantially in the next week or two as we sort out a sane and generalised Atlas dependency installation.

On Fri, Jul 19, 2013 at 2:55 AM, Hoang, Thanh <hoangtv@miamioh.edu> wrote:

Hi all,

I have been analyzing my RNA-seq data on mouse tissues. My RNA-data is single-ended and 51 bp in length. I ran TopHat/Cufflink/Cuffdiff to test to differential gene expression 

In the Cuffdiff's output, I got very high RPKM value for some of miRNA and some other short genes ( less than 100bp). These genes are in the top genes with the highest RPKM. I think the RPKM values of these genes are probably  too high to be true.

test_id

gene_id

gene

locus

sample_1

sample_2

status

value_1

value_2

log2(fold_change)

test_stat

p_value

q_value

significant

ENSMUSG00000093077

ENSMUSG00000093077

Mir5105

5:146231229-146302874

Epithelium

Fiber

OK

1.53E+06

445558

-1.78097

-355.367

0.00715

0.016986

yes

ENSMUSG00000093098

ENSMUSG00000093098

Gm22641

7:130162450-133124354

Epithelium

Fiber

OK

87894.1

36474.7

-1.26887

-0.59863

0.4913

0.587174

no

ENSMUSG00000089855

ENSMUSG00000089855

Gm15662

10:105187662-105583874

Epithelium

Fiber

OK

42868.9

21566.5

-0.99114

-20.7066

0.0186

0.039568

yes

ENSMUSG00000092984

ENSMUSG00000092984

Mir5115

2:73012853-73012927

Epithelium

Fiber

OK

21104.8

8317.49

-1.34335

-447.314

0.0001

0.000354

yes

ENSMUSG00000086324

ENSMUSG00000086324

Gm15564

16:35926510-36037131

Epithelium

Fiber

OK

6443.35

3664.15

-0.81433

-1.52095

0.2129

0.301429

no

ENSMUSG00000092981

ENSMUSG00000092981

Mir5125

17:23803186-23824739

Epithelium

Fiber

OK

5974.14

2390.75

-1.32127

-0.34111

0.5746

0.661937

no

 I checked some forums and they said that this is the drawback of TopHat/Cufflink/Cuffdiff when dealing with short genes. But I am still not so clear about this. Anyone got the same problem? What can I do with this situation?

Anyone suggests any other good tools to test for (1) differential gene expression OR (2) both differential gene expression and gene discovery?

Thank you

Thanh