Annotation file influence in the discovery of non-annotated Fusions

[Bsguerraa]

Hello, thanks for your time and patience! I just want you to know that STAR is amazing! I've been working with Chimeras in RNA-seq in my Ph.D. and I have questions that I hope you can help to give me some insights:

• I'm trying to identify new chimeras from non-cancer tissues, I have plenty of RNA-seq data PE 100bp each and I want to know the influence of the annotations files (.GTF) in the exclusion of possible true candidates non-annotated yet. Does STAR or STAR-Fusion exclude this type of read? I think is true that the discovery of junction-reads in splice-aware aligners is compromised by the lack of annotation.

• I'm trying to run STAR with a different combination of parameters without the annotation, trying to boost the junction read discovery, like TwoPass modes. How do you think this should influence the STAR-Fusion pipe-line? Also, the --NonGTAGScore set to 0 since I don't want to limit the possibility of discard noncanonical splicing sites.

• What kind of Parameters for STAR and Filters in STAR-Fusion do you think is a good idea to change to address my objective? I know that in my research the number of false-positive going to be high, maybe there isn't much I can do since I'm trying something new. Right now I'm using: --outFilterMultimapNmax 100 \ --peOverlapNbasesMin 10 \ --alignSplicedMateMapLminOverLmate 0.2 \ --alignSJstitchMismatchNmax -1 -1 -1 -1 \ --chimSegmentMin 10 \ --chimJunctionOverhangMin 10 \ --chimScoreDropMax 10 \ --chimScoreJunctionNonGTAG 0 \ --chimScoreSeparation 1 \ --chimSegmentReadGapMax 3 \ --chimMultimapNmax 50 \ --outFilterMismatchNmax 33 \ --alignIntronMax 3000000 \ --alignMatesGapMax 3000000 \ --chimMainSegmentMultNmax 10 \

Thanks!

[alexdobin]

Hi @Bsguerraa

you may also want to ask this question on STAR-Fusion site: https://github.com/STAR-Fusion/STAR-Fusion

My thoughts: STAR-Fusion only works with fusions between annotated genes. STAR can detect chimeras between unannotated loci - however, most of them are likely false positives. STAR-Fusion needs to do a lot of careful filtering to bring the false-positive rates down. You may want to check its filtering algorithms, maybe you can repurpose them for your use-case.

The STAR parameters that STAR-Fusion uses are listed here: https://github.com/STAR-Fusion/STAR-Fusion/wiki#KickstartMode They would be a good starting point.

Cheers Alex