CuffDiff 0 FPKM Output/Incorrectly Identified Differential Expression?

[cw11]

I’m having some problems with CuffDiff – or rather, the output I am getting is completely baffling to me. I believe the program is incorrectly identifying genes as differentially expressed due to an error.

I have three conditions that I am testing – we’ll call them GG, AG, and AA. For Condition GG, I have 2 samples, for Condition AG, I have 3 samples, and for Condition AA, I have 5 samples. When I run CuffDiff, I get a list of differentially expressed genes with q-values less than 0.05. However, when I actually examine the FPKMs for my genes, I see results like this:

GG: 0 (OK)
AG: 11.6888 (OK)
AA: 10.7249 (OK)

In this instance, CuffDiff reported GG as being differentially expressed relative to AG, and GG as being differentially expressed relative to AA.

Viewing my reads in IGV, however, I see MANY reads for this gene in ALL conditions and samples. I cannot understand why my FPKMs are 0. I have also calculated the RPKM values for this gene’s transcripts using SeqGene, as a “quality control” of sorts. When I do this, all RPKMs for this gene are > 1; a representative transcript has RPKM values like this:

GG: 2.31, 2.04
AG: 2.63, 2.59, 2.5
AA: 2.32, 2.15, 2.42, 2.33, 3.02

For my pipeline, I am usingTophat to map reads and find splice junctions, followed by Cufflinks to find novel transcripts, CuffMerge to merge these novel transcripts together with a reference genome as guide, and CuffDiff to identify differences in transcript expression. For reference, here is a summary of the calls I’ve made (from most to least recent).

CuffDiff:

bsub -P Lithium -q iweek -R rusage[mem=8] -R span[hosts=1] -n 4 -o /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/07_CuffDiff_10.19.2011_01/output.txt cuffdiff -o /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/07_CuffDiff_10.19.2011_01 -p 12 -L AG,GG,AA /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/06_CuffMerge_10.19.2011_02/merged.gtf /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_TGTGTG_1/accepted_hits.bam,/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_GAGAGA_1/accepted_hits.bam,/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_ACACAC_1/accepted_hits.bam /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_CTCTCT_1/accepted_hits.bam,/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_TGTGGA_1/accepted_hits.bam /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_GAGAAC_1/accepted_hits.bam,/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_ACACCT_1/accepted_hits.bam,/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_CTCTTG_1/accepted_hits.bam,/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_CTTGTG_1/accepted_hits.bam,/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_TGGAGA_1/accepted_hits.bam

CuffMerge:

cuffmerge -o /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/06_CuffMerge_10.19.2011_02 -g /psych/neuro/people/wright/Project_3_RNASeq/Homo_sapiens/UCSC/hg19/Annotation/Genes/genes.gtf -s /psych/neuro/people/wright/Project_3_RNASeq/Homo_sapiens/UCSC/hg19/Sequence/WholeGenomeFasta/genome.fa /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/02_DeMultiplexingCode/F_02_FileForTestMerging.txt

Where FileForTestMerging looked like this:

/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_ACACAC_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_ACACCT_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_CTCTCT_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_CTCTTG_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_CTTGTG_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_GAGAAC_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_GAGAGA_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_TGGAGA_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_TGTGGA_1/transcripts.gtf
/psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_TGTGTG_1/transcripts.gtf

Cufflinks:

bsub -P Lithium -q iweek -o /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_ACACAC_1/output.txt cufflinks -o /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/04_Cufflinks_10.18.2011_09/FastQDemultiplexed_ACACAC_1 -g /psych/neuro/people/wright/Project_3_RNASeq/Homo_sapiens/UCSC/hg19/Annotation/Genes/genes.gtf /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_ACACAC_1/accepted_hits.bam

Tophat (one of 10 calls):

bsub -P Lithium -q iweek -R rusage[mem=8] -R span[hosts=1] -n 4 -o /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_CTTGTG_1.txt tophat -o /psych/neuro/people/wright/Project_3_RNASeq/01_RNASeq_Analysis/03_TophatTest_10.12.2011_11/FastQDemultiplexed_CTTGTG_1 -p 4 -a 5 -F 0.0 -r 100 --segment-length 20 /psych/neuro/people/wright/Project_3_RNASeq/hg19-male /psych/inf/cwright/05_DemultiplexedFastQFiles/FastQDemultiplexed_CTTGTG_1.fq /psych/inf/cwright/05_DemultiplexedFastQFiles/FastQDemultiplexed_CTTGTG_2.fq

Any thoughts, help, comments on the CuffDiff output (or suggestions on the pipeline, for that matter) would be greatly appreciated!

[gringer]

Does your gene have another isoform that could be taking all the reads? Cufflinks considers isoform abundance in its FPKM calculations, so that might have something to do about it.

[lmf_bill]

I got the similar problems. For some genes without isoforms, cufflinks or cuffdiff estimate 0 FPKM, but the real case is they are expressed high.

[cw11]

Thanks! What has everyone doing to deal with these false positives? Just filtering out genes which have FPKMs of 0 for one of the conditions?

(I also understand from a different thread - http://seqanswers.com/forums/showthread.php?t=14864 - that the tuxedo group is planning to address some of these issues in their upcoming version 1.2.0... anyone know when this is expected to come out?)

[pengchy]

The problem still exist for cufflinks version 2.2.1. I checked One example that one gene without any reads hit, be reported FPKM 0. However, not all genes without any reads hit were reported. How to explain?