TOPHAT2 vs RNA-STAR 2X the mapped %.

[tonup69]

Hello all. I have been working with some very clean RNAseq data generated from Drosophila and have come across a few issues. First of all, I am using our local install of GALAXY - forgive me, but I am only a part-time bioinformaticist! That being said, I have used this interface for about five years now, so I am comfortable. The main issue troubling me right now is that I can use TOPHAT2 for mapping and am able to generate differential gene expression output using the Bowtie/Cufflinks/Cuffmerge/Cuffdiff pipeline. The problem is that the mapping is VERY low (about 50-60% for any of the fastqsanger files). I am only using clean data trimmed to 200bp or less (Ion Proton) with quality >20 median score. The problem is, when I use RNA-STAR I get multiple issues in the pipeline (problems with cufflinks that end up being problems in Cuffdiff that I don't want to get into here). What I want to know is, why do I get >80% mapped reads (and higher) when I use RNA-STAR, but only about 50% mapped reads with TOPHAT2? Is there some reason for this?

Thanks in advance. Working through this....

[Brian Bushnell]

TopHat is not very tolerant of errors in data. I'd recommend avoiding all of the Tuxedo pipeline where possible; I've always found it to be slow and unstable. Deseq and Edger seem to give more accurate results, anyway.

[tonup69]

Brian - I am just talking about mapping right now. Even if I switched to Deseq or Edger (which I could do on our local install) I still need a gaped aligner to map the sequence with first. Do you use RNA-STAR?

[tonup69]

Here are the actual numbers on the same fastqsanger file (trimmed to 200bp QS>20).

Tophat2:
Reads:
Input: 39248980
Mapped: 20971700 (53.4% of input)
of these: 3246961 (15.5%) have multiple alignments (5659 have >20)
53.4% overall read alignment rate.

RNA-STAR:
Started job on | Feb 15 12:26:35
Started mapping on | Feb 15 12:26:38
Finished on | Feb 15 12:35:13
Mapping speed, Million of reads per hour | 274.36

Number of input reads | 39248980
Average input read length | 166
UNIQUE READS:
Uniquely mapped reads number | 32172637
Uniquely mapped reads % | 81.97%
Average mapped length | 164.84
Number of splices: Total | 7873494
Number of splices: Annotated (sjdb) | 0
Number of splices: GT/AG | 7824667
Number of splices: GC/AG | 46067
Number of splices: AT/AC | 2760
Number of splices: Non-canonical | 0
Mismatch rate per base, % | 0.47%
Deletion rate per base | 0.14%
Deletion average length | 1.17
Insertion rate per base | 0.24%
Insertion average length | 1.15
MULTI-MAPPING READS:
Number of reads mapped to multiple loci | 5162521
% of reads mapped to multiple loci | 13.15%
Number of reads mapped to too many loci | 59135
% of reads mapped to too many loci | 0.15%
UNMAPPED READS:
% of reads unmapped: too many mismatches | 0.00%
% of reads unmapped: too short | 4.42%
% of reads unmapped: other | 0.30%

[Brian Bushnell]

Brian - I am just talking about mapping right now. Even if I switched to Deseq or Edger (which I could do on our local install) I still need a gaped aligner to map the sequence with first. Do you use RNA-STAR?

Nope, I use BBMap I have heard good things about STAR, but I've never benchmarked it.

[Michael.Ante]

Hi Tonup69,

check your alignments with rseqc, especially the clipping profile will be interesting.
Maybe your alignment rate with STAR is very good, but the alignment-length might be quite short...

[tonup69]

Well, its a nice thought, but we don't have either of those wrappers on our local install and I am not the admin for the box. I will bring it up, but I am limited to what we have on our server.