Aligner for PacBio RNASeq

[RamP]

I am wondering if I can use BLASR for aligning RNASeq reads from Human cell lines. I am interested in identifying the splice sites. I read this paper "Evaluation of tools for long read RNA-seq splice-aware alignment" (Bioinformatics, 2018) where different splice aware tools were compared and concluded GMAP to be the best among the compared ones. However, they didn't include BLASR in their evaluation. Could the reason be that BLASR is suitable only for genomic read alignments? Thanks.

[Magdoll]

You cannot use BLASR. You can use minimap2, GMAP, STAR, etc.
A tutorial for these aligners for PacBio RNA (Iso-Seq) data is available here:

https://github.com/Magdoll/cDNA_Cupcake/wiki/Best-practice-for-aligning-Iso-Seq-to-reference-genome:-minimap2,-GMAP,-STAR,-BLAT

[RamP]

You cannot use BLASR.

Thanks for your response and the link. What could be the reason for BLASR being not a suitable aligner? I mean, it works like BLAST finding local alignments after incorporating the PacBio error model and that should be good enough to align exon regions on the genome. What am I missing here? Thanks.

[Magdoll]

BLASR is not splice aware - it cannot map exon-exons. Newer aligners like minimap2 are well equipped to handle errors for PacBio data. For most PacBio RNA-seq (what we call Iso-Seq) data, the error rate would be very low anyway.

In more practical terms, BLASR is getting phased out. We encourage users to switch to newer aligners like minimap2 that are being actively maintained and developed.

[RamP]

Thanks for taking time explaining this. One last thing.
Would Minimap2 be effective enough aligning to non-canonical coding regions, for example, fusion transcripts?
Perhaps, I should give it a try myself, but thought I would ask anyway.
Thanks again.

[Magdoll]

Yes both GMAP and minimap2 could work for fusion detection.

In GMAP it would be using the `-n 0` parameter which says "only output a single alignment *unless* the best alignment is chimeric" which essentially screens for fusion transcripts when you look for sequences with 2+ alignment results.

In minimap2 I think it's `--secondary=no`, though admittedly I have tested it less.

I have a follow up script that processes the SAM files to look for fusion: https://github.com/Magdoll/cDNA_Cupc...ng-step#fusion

If you have problems using the scripts above, please file through GitHub Issues.