STAR Alignment Score reduced after trimming

Question

[dec986]

Hello,

I'm comparing three different QC tools (trimgalore, trimmomatic, & bbduk) and then seeing how STAR alignments are affected by all of them, comparing against a noQC run. Specifically, I am comparing the "AS" and "nM" fields in the BAM file. I figured that the alignment scores should increase, but the QC file without any QC gets the best alignment scores, albeit with higher mismatch rates.

I am concerned that the alignments after QC may be worse because of this, specifically the means of the alignment scores can be seen:

trimgalore  - noQC  = -5.57
trimmomatic - noQC  = -2.99
bbduk       - noQC  = -1.15

Why are the QC tools causing worse alignment scores in STAR?

Answer 1

[Devon Ryan]

There's little point in trimming reads with STAR (or other tools that do local alignment), at most trim adapters and very low quality bases (phred scores up to ~3 or so). Presuming you did quality trimming, it's likely that the score you used was much too high, which decreased the number of bases that STAR would have otherwise aligned and therefore decreased the alignment scores. Even adapter trimming isn't perfect, since inevitably you'll end up trimming off a bit of sequence. For both our command line and Galaxy RNAseq workflows, we skip trimming entirely and use STAR with the raw reads. That tends to produce quite nice results.

[dec986]

thank you! who is "our"?

[GenoMax]

who is "our"?

@Devon is currently a bioinformatician/data manager at the Max Planck Institute for Immunobiology and Epigenetics in Freiburg, Germany.

[dec986]

great, thanks, I have to let my colleagues know. We're comparing several different RNA-Seq pipelines.

[Kevin Blighe]

Evidently a great manager too

[Devon Ryan]

Thankfully I manage the data and not the people :)

[dec986]

Is there any official documentation for this? I can't find anything in google searches or in the STAR manual

[Devon Ryan]

That's not going to be documented since it's not a direct result of a tool, but rather an interaction between how alignments work generally and trimming generally changes thing.