SAM result format required in cufflinks

[arthur.yxt]

Hi all,

Tophat is a great, but it seems not working so well with Cufflinks. As Cufflinks require the .SAM result containing the "XS:A:±" field and Tophat don't provide this, I will have to add this field to the end of each line of .sam file.

But after I did this, cufflinks warned me with "XA attribute not supported". And "XA" attribute is there from Tophat. Should I remove "XA" field or put "XS" field in the position of "XA"?

Thanks!

[mmanrique]

Hi Arthur,

have you tried getting .SAM files using Bowtie (http://bowtie-bio.sourceforge.net/index.shtml) instead of Tophat? I think it would work although I haven't tried it yet.

[calliopsis]

cufflinks error

Hi~

I tried the new version bowtie-0.12.1 to get .sam result, and then use it as the input file of Cufflinks, there is the same error: XA attribute not supported, how can I fix it?

[staylor]

XA attribute not supported cufflinks

I get the same error message with SAM produced by bowtie version 0.11.3.

Can you just do a search and replace on the file as suggested by calliopsis?

[staylor]

The way round this seems to be to get a sorted SAM file and remove the offending columns.

I converted my SAM file to a BAM file, sorted the BAM file using samtools, and then with the resulting all.sorted.bam removed the offending columns. Thus:

samtools view all.sorted.bam | cut -f1-11 > all.sorted.sam

Then cufflinks will process this without error.

[Cole Trapnell]

Hi all,

Tophat is a great, but it seems not working so well with Cufflinks. As Cufflinks require the .SAM result containing the "XS:A:±" field and Tophat don't provide this, I will have to add this field to the end of each line of .sam file.

But after I did this, cufflinks warned me with "XA attribute not supported". And "XA" attribute is there from Tophat. Should I remove "XA" field or put "XS" field in the position of "XA"?

Thanks!

Which versions of TopHat, Cufflinks, and Bowtie are you working with? As of TopHat 1.0.11, all spliced alignments should have the XS "source strand" attribute attached. As of Cufflinks 0.8.0, SAM attributes other than XS and NM are silently ignored, and should produce no warnings.

[staylor]

0.11.3.

Will doing the cut affect the accuracy of the cufflinks calls?

[Cole Trapnell]

0.11.3.

Will doing the cut affect the accuracy of the cufflinks calls?

Definitely. Without those attributes, Cufflinks can't place transfrags on a strand, and if you are assembling transcripts (i.e. not using an annotation), it will merge genes that overlap on opposite strands. Even if you are using an annotation, the quantitation accuracy could be reduced. You should do your best to feed Cufflinks spliced alignments that are stranded with the XS. The upcoming version 0.8.2 will complain loudly when you omit them.

[staylor]

Definitely. Without those attributes, Cufflinks can't place transfrags on a strand, and if you are assembling transcripts (i.e. not using an annotation), it will merge genes that overlap on opposite strands. Even if you are using an annotation, the quantitation accuracy could be reduced. You should do your best to feed Cufflinks spliced alignments that are stranded with the XS. The upcoming version 0.8.2 will complain loudly when you omit them.

Ok I will upgrade bowtie asap.

In the meantime is there a simple way I can process the SAM/bowtie output to be compatible with this version of cufflinks or do I just have to run the searches again?

[thinkRNA]

You should do your best to feed Cufflinks spliced alignments that are stranded with the XS. The upcoming version 0.8.2 will complain loudly when you omit them.

Can one get bowtie to output strand information? I don't want to use tophat as of now because I am only interested in differential expression and not splicing. However, I want to make sure my reads are getting aligned to the correct strand and that cufflinks is taking this into consideration.

[Cole Trapnell]

If the reads are strand specific, then you can simply run the forward reads against the forward strand, and attach the XS:A:+ attribute ex post facto, and similarly against the reverse strand.

However, I strongly recommend NOT using Bowtie alignments with Cufflinks to calculate differential expression, even at the gene level. Cufflinks' calculates expression for genes by adding the relative abundances of the individual transcripts, and these are vastly more difficult to calculate without spliced reads. I urge you to use TopHat, even with novel junction discovery disabled if needed, in order to align your reads.

[thinkRNA]

Thanks so much for your prompt reply. I will begin using tophat spliced alignment now. I would like to "customize" my tophat runs with different bowtie parameters. I think someone in this forum has suggested how to do this (by editing the tophat python script), but it would be helpful as an option in tophat. Thanks a million!

[Pejman]

However, I strongly recommend NOT using Bowtie alignments with Cufflinks to calculate differential expression, even at the gene level. Cufflinks' calculates expression for genes by adding the relative abundances of the individual transcripts, and these are vastly more difficult to calculate without spliced reads. I urge you to use TopHat, even with novel junction discovery disabled if needed, in order to align your reads.

Dear Cole
Could you please elaborate how much of a catastrophe would that be to use Bowtie -> Cufflinks instead of Tophat -> Cufflinks, particularly in case of no-new-junctions?

I'm using single end 30bp SOLiD RNAseq colorspace reads from Human samples, which does not quite fit to Tophat defaults. I'm using both tophat and cufflinks only with given known junctions. The problem is that with Tophat I get around 5 times less sequences aligned, compared to Bowtie, and there is not much flexibility in Tophat to manipulate the alignment settings.

[zee]

So according to Cole (Cufflinks author) we shouldn't be using Cufflinks with SAM results from other alignments.
If I am using BWA to map my reads I am going to potentially lose out because there are no XS tags in my SAM format.

[Pejman]

So according to Cole (Cufflinks author) we shouldn't be using Cufflinks with SAM results from other alignments.
If I am using BWA to map my reads I am going to potentially lose out because there are no XS tags in my SAM format.

Well, missing tags are not essential problem, cuz you can't run cufflinks without them With bowtie you can easily make a compatible format, the question is how much sense it makes to do so. The problem with generic alignment programs like Bowtie is that they essentially do not take splice junctions into account while it is the core idea in Cufflinks expression estimation. well, at least thats my impression!