tuning the tophat/cufflink pipeline

[oliviera]

Dear all,

I am using tophat/cufflinks on a set of single end sequencing data from 4 different biological samples (without replicates). My first goal is to compare the rpkm in the different conditions.

Here is how I use tophat:
tophat -p 4 -G /path/to/gff_file /path/to/genome_file /path/to_input_file

And here cufflinks usage I used:
cufflinks -G /path/to_gff_file /path/to/sam_file

When I look to the transcript.expr file I am surprize to see that sometimes same ensembl transcript have multiple rpkm, such:

ENSDART00000000198 1367346 Zv8_scaffold3091 65836 148200 0.03 1 1 0 0.1 0.02 2139
ENSDART00000000198 1367332 Zv8_scaffold3091 65836 148200 1.47 1 1 1.03 1.9 0.77 2139

This is true also in the gene.expr file...

Is it because I didnt use the -g 1 option in tophat to restrict to single hit in the genome?
Could help me to tune the options in tophat and cufflink to avoid this splitting of rpkm from the same location?

Cheers

Oliviera

[sjm]

advice on tophat/cufflinks genes/transcripts and RPKM

I've seen similar behavior when using cufflinks to look at RPKM assigned to Ensembl transcript entries. If you look carefully at exactly which chromosomal locations are involved for each transcript entry, I think you'll find that each line in transcripts.expr (or genes.expr) corresponds to a different chromosomal location within the same ENS transcript entry.

i.e. I don't think cufflinks is outputting multiple RPKMs for the same 'bundle' of reads - it's just that these 'bundles' match to multiple areas of a given ENS entry.

I get around this by having tophat and cufflinks do RPKM calculations for the whole ENSG genes rather than individual ENST transcripts. Tophat with no -G, but use it in cufflinks, e.g.:

tophat -o /path/to/output_folder /path/to/bowtie_index input_reads/fastq
cd /path/to/output_folder
cufflinks -G Ensembl.gtf accepted_hits.sam

Then extract your data from the genes.expr file.

Hope that helps!

[oliviera]

Dear sjm,

Thanx for your answer. I will try to do as you say. I am not sure of it, but I thought that providing the gff file to tophat will help the mapping of the reads, that s why I thought to provide the gff file in tophat. Is it indeed the case?

Here are the first lines I got with cufflinks: (the chromosome nb start and end are identical...)
gene_id bundle_id chr left right FPKM FPKM_conf_lo FPKM_conf_hi
ENSDARG00000000189 1367344 Zv8_scaffold3091 65836 148200 0.03 0 0.1
ENSDARG00000000189 1367346 Zv8_scaffold3091 65836 148200 0.03 0 0.1
ENSDARG00000000189 1367388 Zv8_scaffold3091 65836 148200 0.32 0.12 0.52
ENSDARG00000000189 1367400 Zv8_scaffold3091 65836 148200 0.29 0.1 0.48
ENSDARG00000000189 1367408 Zv8_scaffold3091 65836 148200 0.54 0.28 0.81
ENSDARG00000000189 1367410 Zv8_scaffold3091 65836 148200 0.38 0.16 0.6
ENSDARG00000000189 1367412 Zv8_scaffold3091 65836 148200 0.89 0.56 1.23
ENSDARG00000000189 1367414 Zv8_scaffold3091 65836 148200 0.54 0.28 0.81
ENSDARG00000000189 1367416 Zv8_scaffold3091 65836 148200 0.77 0.45 1.08
ENSDARG00000000189 1367418 Zv8_scaffold3091 65836 148200 0.48 0.23 0.73
ENSDARG00000000189 1367420 Zv8_scaffold3091 65836 148200 0.16 0.02 0.3
ENSDARG00000000189 1367424 Zv8_scaffold3091 65836 148200 0.06 0 0.15
ENSDARG00000001712 1367600 Zv8_scaffold3094 66975 111170 0.14 0 0.33

I realised that this is happening almost exclusively when the transcripts are mapped to the scaffold genomic location (ie the genome I use is not as good as the mouse and as still some parts which are not assembled into chromosome).

Cheers

Oliviera