Mapping using Tophat2 in different species

[lischoco]

Hello, all

I am a new person in this world and would like your help.

I am trying to analyze Illumina HiSeq2000 results from a nonmodel, rodent both by de novo and genome-based analysis. I am waiting until the de novo results, and meanwhile am trying to map the rodent sequence results to the mouse genome.

I have done the mapping using default settings in Tophat2, but I saw that not much sequences were mapped to the genome, e.g. the accepted_hits.bam were 33.5MB compared to 2.01GB for unmapped.bam for one of my samples.

I have thought possible reasons may be 1) low quality reads getting in the way 2) adapter sequences getting in the way 3) the tophat2 settings were too stringent to compare the two species.

As for 1), when I saw the results from FastQC, I didn't see any problems in the quality of the reads (they were in the green area). As for 2), there were no sequences that were overrepresented, although there were a few k-mers that were enriched in the first 10 bases.

So, I was suspecting that the main culprit was 3) the mapping conditions being too stringent, and I read a few threads here that said that changing a few parameters in bowtie2 may help. I tried mapping the same sequencing results changing the first time, N:0 <- 1, and the second time, N:0 <- 1 & L:20 <- 10, but still had no luck.

So, I'd like to ask, if there are any hints as to what I should change in order to improve mapping/make the threshold more lenient? Specific numbers and variables, published papers with parameters are greatly appreciated!

Thank you in advance!
& Sorry for the long question!

[Brian Bushnell]

I suggest you use BBMap; it is much more tolerant of errors than Tophat, and thus better for cross-species mapping. For vertebrate RNA-seq, you should include these flags:
xstag=us intronlen=10 maxindel=200000 ambig=random

You can increase sensitivity with the 'minratio' flag. Default is 0.56; for cross species RNA-seq to a genome, you may want to set it as low as 0.2 (minratio=0.2). Note that the 'xstag=us' flag is for unstranded RNA-seq data; if you know that your data is 'firststrand' or 'secondstrand' then set 'fs' or 'ss' instead of 'us'.

Ganbatte!

[lischoco]

Dear Brian

Thank you for your quick reply! I greatly appreciate it!
I will try out BBMap. I might ask for help when I get stuck.

Just out of curiosity:
I see that many people are using tophat for their genome-based alignment, but if tophat is not that tolerant towards errors, does that mean the species on which these people are focusing are genome-known species/really close relatives?

Thank you!

[Brian Bushnell]

lischoco,

Tophat has been available much longer, and is thus better known, which is why it is more commonly used. And typically, RNA-seq alignment is not done to different species, even close relatives - it's almost always done to the genome of the species of interest. Getting robust and consistent differential expression data already difficult without adding additional complications.

-Brian