BWA with Cufflinks (instead of TopHat)

[vinay052003]

I have a basic question (may be very naive as well). How well Cufflinks perofrms if I use BWA alignment for my RNA-Seq data instead of TopHat?
Since the output from both the programs (BWA and TopHat) is in bam/sam format that is what Cufflinks required, any aligner can be used in my opinion.
But according to Cufflinks home page, they recommend TopHat for Cufflinks.
So I am bit suspicious about using BWA instead of TopHat.
Any thoughts or experiences with the same?

Thanks.

[arvid]

Never tried BWA, but GSNAP works well with Cufflinks.

[kopi-o]

I don't see there being much point in using BWA + Cufflinks. BWA doesn't do split alignment like TopHat (to handle spliced transcripts), so it only makes sense to use BWA to map directly against transcripts. But Cufflinks is designed (if we are just talking about its quantification mode rather than the assembly mode) to estimate abundances of different splice isoforms based on a genomic transcript annotation, and importantly, spliced alignments from TopHat where the CIGAR string in the BAM/SAM file contains "N" characters.

[lukas1848]

BWA doesn't do split alignment like TopHat (to handle spliced transcripts), so it only makes sense to use BWA to map directly against transcripts.

Are you sure that bwa doesn't do spliced alignments?
Another reason for choosing bwa over tophat would be that bwa supports longer reads (like 454), which tophat does not. At least to my knowledge...

[kopi-o]

Not as far as I know. It does permit (short) gaps in the alignments, which is a different thing.

[wokai001]

The main issue in using Tophat is that you can get gapped alignments (which span Intron-Exon junctions). Tophat does that by splitting reads and mapping the fragments independently.

Bowtie and BWA 's gaps usually are only a handful positions long. Their Alignment is easier done and for that, they are faster than tophat.

You use Tophat and cufflinks to get Expression estimates for differen Isoforms of your genes.

Combining first Bowtie or BWA followed by Tophat can make sense: You first filter out your ungapped aligns and then align the "unmapped" Reads in a second step with Tophat. I tried this out and found about 95% of former BWA unmapped reads as gapped (i.e. crossing an Intron-Exon border) Alignments in Tophat.

[ferrari]

How can you run tophat after BWA?
Tophat needs fastQ, is SAM/BAM possible?

[chadn737]

How can you run tophat after BWA?
Tophat needs fastQ, is SAM/BAM possible?

You can't they are two different tools.

[wokai001]

Run tophat after bwa

How can you run tophat after BWA?
Tophat needs fastQ, is SAM/BAM possible?

Run bwa until you have a sam file:

bwa aln ...
bwa samse ...

Extract unmapped reads from sam-file into bam-file:
samtools view -f 0x4 -b -S -o unmap.bam infile.sam

Tophat has a tool which converts bam into fastq:
bam2fastx --fastq unmap.bam > umap.fastq

Then proceed with tophat as usual on the new unmap fastq.

With RNA-seq, you can filter out ungapped reads with this procedure.