cuffdif on different FPKM

[drunk_coder]

hi,all
Recently, I used tophat and cufflinks to find out different gene expression within three RNA-seq library.

I use cuffdif to analyze the different gene expression within sample1 and sample2, and then sample1 and sample3.

when I compare the gene_ex.diff file of sample1 and sample2 with the gene_ex.diff file of sample1 and sample3, I find the the same gene's FPKM of sample1 in this two files is different. I feel confused about that.

then I find that the every gene's sample1's FPKM in sample 1_2 divide sample1's FPKM in sample 1_3 is a constant

Is the normalize of library cause the problem?
I find the read_groups.info file in the output of cuffdif
this is sampe 1_2
na_tophat/accepted_hits.bam q1 0 1.61963e+07 1.30384e+07 1.84746 1
ZM/accepted_hits.bam q2 0 7.76888e+06 1.30384e+07 0.57735 1

this is sample 1_3

na_tophat/accepted_hits.bam q1 0 1.61963e+07 1.35408e+07 2.52492 1
PM/accepted_hits.bam q2 0 6.66703e+06 1.35408e+07 0.411644 1

If so, why should do this normalization? why not use the raw FPKM ?

Hope

[NicoBxl]

which version of tophat and cufflinks ?

Is the fpkm of a gene the same in the cufflinks output and in the cuffdiff output ?

[drunk_coder]

which version of tophat and cufflinks ?

Is the fpkm of a gene the same in the cufflinks output and in the cuffdiff output ?

I used tophat v2.0.3 and cufflinks v2.0.0

Because I just want to known the different expression in gene level, I didn't use cufflinks to transcript asembly, so I didn't get the cufflinks FPKM.

I find that cuffdiff do normalize the data based on the library size, so cause gene1 FPKM in sample1 of sample 1_2 is different with gene1 FPKM in sample1 of sample 1_3.

THis makes me have to do cuffdiff again, if I want to compare sample2 and sample3 roughly.

[NicoBxl]

I think using DESeq or edgeR (two R bioconductor package) will be more efficient for you if you have an annotation file.

First align with tophat
Then use htseq-count to extract the read count for each feature (in your case, for each gene)
Then use DESeq or edgeR to compute DE analysis.
DESeq and edgeR can normalize each sample, and you can after perform clustering on it.

[hpimentel]

hi,all
Recently, I used tophat and cufflinks to find out different gene expression within three RNA-seq library.

I use cuffdif to analyze the different gene expression within sample1 and sample2, and then sample1 and sample3.

when I compare the gene_ex.diff file of sample1 and sample2 with the gene_ex.diff file of sample1 and sample3, I find the the same gene's FPKM of sample1 in this two files is different. I feel confused about that.

then I find that the every gene's sample1's FPKM in sample 1_2 divide sample1's FPKM in sample 1_3 is a constant

Is the normalize of library cause the problem?
I find the read_groups.info file in the output of cuffdif
this is sampe 1_2
na_tophat/accepted_hits.bam q1 0 1.61963e+07 1.30384e+07 1.84746 1
ZM/accepted_hits.bam q2 0 7.76888e+06 1.30384e+07 0.57735 1

this is sample 1_3

na_tophat/accepted_hits.bam q1 0 1.61963e+07 1.35408e+07 2.52492 1
PM/accepted_hits.bam q2 0 6.66703e+06 1.35408e+07 0.411644 1

If so, why should do this normalization? why not use the raw FPKM ?

Hope

Hi. Can you please post your Cuffdiff command?

[drunk_coder]

Hi. Can you please post your Cuffdiff command?

Thank you for your reply

sample1_2(NTA and PM two samples)
command : cuffdiff -o NTA_PM -b Human_66.fa -p 8 -u Homo_sapiens.GRCh37.66.gtf ./nta_tophat/accepted_hits.bam ./PM/accepted_hits.bam

sample1_3 (NTA and ZM two samples)
command : cuffdiff -o NTA_ZM -b Human_66.fa -p 8 -u Homo_sapiens.GRCh37.66.gtf ./nta_tophat/accepted_hits.bam ./ZM/accepted_hits.bam