Just curious- for #5 here, do the riboseq reads get aligned with the novel transcripts supplemented alongside standard annotations (ie genome + "hybrid" GTF)? or is the "hybrid" reference only supplied post-alignment for ORF calling?

To be clear, others here are the ones making the methodological calls, but FWIW it's a bit different per caller. For Ribo-TISH, Ribotricer, Rp-Bp and PRICE: no re-alignment. They reuse the existing genome BAMs from the normal single STAR pass and just get pointed at the hybrid GTF for annotation. That works because they operate on genome coordinates. RiboCode is the exception: it needs a transcriptome-coordinate BAM, and that's index-specific, so when

--extended_orf_analysis

is on we do a genuine second pass. It's a fresh STAR genome index built from the genome FASTA + hybrid GTF (via

--sjdbGTFfile

), re-aligning from the raw Ribo-seq FASTQs (not remapping the existing BAMs), plus a hybrid transcriptome FASTA for the

--quantMode TranscriptomeSAM

step RiboCode needs. Note it's Ribo-seq reads only in that second pass, not RNA-seq. (RNA-seq gets its own separate hybrid re-alignment, but only for the DTE denominator when

--contrasts

is supplied.)

Thanks- so lets talk through a hypothetical to make sure i follow. Let's say RNAseq supports a novel transcript not at all represented in the standard GENCODE annotation. After stringtie, that would find its way into the hybrid GTF. A riboseq read that maps to that novel transcript should (?) align all the same whether that transcript was supplied alongside the reference or not given it is still a genomic alignment. However it's maybe not impossible that if the novel transcript shared some structure to that of a GENCODE-annotated transcript, some riboseq reads could get misallocated, though I'm not sure if that is true. Wouldn't it be safer nonetheless to give all riboseq reads (regardless of method) a true 2nd pass on the genome + hybrid GTF?

I think it's important to distinguish between assigning reads to unique genomic loci and determining which transcript they originated from. Providing STAR with a hybrid GTF can help with the former, but not necessarily the latter. The main benefit would be for reads spanning newly annotated splice junctions, which Ribo-seq is already relatively poor at resolving given the short read lengths. Adding more junctions can also introduce competing alignments and potentially reduce the proportion of uniquely mapped reads, so we'd want to be reasonably confident in the novel junctions we're introducing. I'm not convinced a second alignment would substantially improve transcript assignment, but it would be interesting to quantify how many additional junction-spanning reads it actually recovers.

thanks- so either way, you're reasonably confident that a novel previously-unannotated ORF supported by both RNA+riboseq data would get reliably called with the existing path?

Unless the candidate ORF depends on reads spanning a novel splice junction, I think it should have just as good a chance of being called, provided the novel transcript is considered by the callers. That said, I wouldn't necessarily trust any individual caller to reliably identify every translon (excuse my propagandising). They all have their blind spots, so it's not simply a case of "if it's translated, it'll be called". That's a separate issue from whether we need to realign the reads.

re: that dependency, riboseq reads that align to novel exons would have no problem, its just reads that would in theory span two novel exons with a region in between presumably spliced may get missed. correct?

The exons should be uniquely mappable in the genomic context but yes. It is a question of being confident in where the exonic sequence that your captured ribosome was translating originated.

got it, thanks for that!

With the provisos Jack cites, it wouldn't be at all difficult to tweak the wiring so that all the callers use the results of the hybrid alignment (with the benefit of the newer splice junctions). We'd just want to be sure we weren't making things worse by affecting the mapping in unexpected ways.