why does Cufflink does not report strand information for unspliced reads

[Chirag]

Hi there,

I have been using cufflink to predict transcript, but cufflink does not provide strand info on it. Could someone please suggest me how to solve that.
This is how i run the program

# Mapping of reads by tophat
tophat --num-threads 7 --solexa-quals --output-dir Transcript_Mapped/Eggs --butterfly-search /export/data/alignment_references/bowtie/danRer7 Raw_Reads/Eggs_fastq


# Construction of trasncripts
cufflinks --output-dir Eggs/ --num-threads 4 --frag-bias-correct /export/data/alignment_references/bowtie/danRer7.fa --multi-read-correct --upper-quartile-norm /home/chirag/Projects/danRer7_RNASeq/Project_FM009/Transcript_Mapped/Eggs/accepted_hits.bam

# Merge transcripts
cuffmerge --num-threads 6 -s /export/data/alignment_references/bowtie/danRer7.fa -o Version-1 assemblies.txt



Transcript.gtf and Merge.gtf is now covered into .bed

chr1 1906 4188 CUFF.1.1 1000 . 0 0 0 1 2282, 0,
chr1 39115 39565 CUFF.2.1 1000 . 0 0 0 1 450, 0,
chr1 39666 39981 CUFF.3.1 1000 . 0 0 0 1 315, 0,
chr1 1 745 CUFF.4.1 1000 . 0 0 0 1 744, 0,
chr1 36295 36812 CUFF.5.1 1000 . 0 0 0 1 517, 0,
chr1 4343 35664 CUFF.6.1 1000 - 0 0 0 4 829,86,65,858, 0,6063,6921,30463,
chr1 4343 27499 CUFF.6.2 1000 - 0 0 0 4 829,86,65,837, 0,6063,6921,22319,


We can see when the transcript is unspliced, it is unable to predict the strand.
Even if i look at the GTf file, there is no strand information.


The RNASeq data i used is 76 bp long, from Illumina, and has not strand information.
As far as i am aware, Cufflink should be able to predict the strand irrespective of the strand specific RNASeq data. I guess, it is not because of low coverage, since we have around 250 million reads per sample.

Could you please suggest me on this, how can this be solved ?
Is there any parameters which i have missed while making the transcripts ?

Thanks for your help in advance !

regards
Chirag

[Simon Anders]

How do you think cufflinks predicts the strand information if it is not in the data?

Of course it does not have magic omniscience. Rather, it looks at splice sites to infer the strand. (The splice donor on the 5' side of an intron is (nearly) always GU, the splice acceptor on the 3' side is AG.) If there is no intron, this will not work.