Possible that ORF annotation is overly greedy

[aleighbrown]

I've been testing out the ORFannoate tool on mis-splicing events after TDP-43 knockdown

I’m highlighting an exon here that I am like 99.999% certain leads to RNA decay through the NMD pathway (RNA levels decrease after TDP-43 KD, if you inhibit NMD with a UPF1 inhibitor, that loss is rescued).

ORFannotate is generating these crazy 5’utrs upstream reading sequences and saying every isoform with that exon is protein-coding, which, seems biologically unlikely to me.

Just wondering if there's some upstream settings I should tweak to get it to be less accepting of extremely long 5'utrs?

[egustavsson]

Hi AL,

Thank you for raising this issue, we are looking into it.

Would you be able to provide us with a small GTF file containing the transcripts shown above? That would help us reproduce the behaviour locally and better understand why those ORFs are being prioritised.

Thanks,

Emil

[aleighbrown]

ORFannotate_annotated.tsv

Sorry for the delay - extension changed from .gtf to .tsv so github would accept the upload, it's just a GTF

[egustavsson]

Thank you again for raising this. After looking into it further, I think what is happening here is that the cryptic exon inclusion is disrupting the canonical coding frame and putting the known start/stop relationship out of frame. ORFannotate (CPAT) is then identifying a shorter downstream ORF as the best remaining coding candidate, which results in most upstream exons being labelled as 5′UTR.

However, from the examples you shared, it also looks like these transcripts contain predicted uORFs and transcript structures that would likely make them sensitive to NMD.

I think this highlights an important limitation in the current prioritisation logic. At the moment, ORFannotate is primarily identifying plausible ORFs, but it is not yet sufficiently considering the biological impact of uORFs. We are going to work on integrating biological logic into ORFannotate so that these cases can be better classified.

Thanks again, this is a very useful example for improving the tool.