Identifying non-surjectable sites in a minigraph-cactus graph

For a graph built with Minigraph-Cactus, is there any way to determine which portions of a non-reference path wouldn't be able to be surjected to the primary reference coordinate system?

As a theoretical example, if I were to take short reads (150-bp, paired, with zero errors) generated from one of the assemblies included in the graph as a non-reference path, align them to the graph with vg giraffe and output a BAM file in the coordinate system of the primary reference, some sites will inevitably be missing (e.g., because the non-reference path includes a sufficiently large alternate allele at a locus). Is there a way to identify such non-surjectable sites in both the primary reference and non-reference paths with respect to one another, just from the graph, rather than proceeding through alignment and variant calling?

• 1,394 views • link

written 1.8 years ago by avril •

1 answer

If you mean PAVs - presence absence variations, you can check out odgi pav with the nice tutorial here :

https://odgi.readthedocs.io/en/latest/rst/tutorials/presence_absence_variants.html

I think minigraph-cactus outputs odgi, if not conversion from GFA format is very easy using odgi.

• 0 views • link

written 1.8 years ago by colindaven

thanks! I have used odgi pav, but didn't think of it as a potential solution here. If I were to use that, I think what I would be looking for is missing sequence in the primary reference relative to each path, and I haven't had much luck setting a non-reference path as a reference for that purpose. I'm also interested in a genome-wide estimate of these types of variants, and I'm not sure that this approach would be efficient for that? (I am working with pretty large and often messy plant genomes.)

Maybe what I'm actually looking for are nodes that are not traversed by the primary reference and that are of sufficient length that surjection for the purpose of calling SNPs would not work. When I call SNPs/MNVs/INDELs from graph-aligned reads, I surject to the primary reference and keep called variants <50 bp, because in theory, variants >=50 bp are caught by vg call in a separate process. So nodes >=50 bp in length that would be fully covered by reads (i.e., that entire variant is present in the reseq sample) would produce non-surjectable alignments? Am I thinking about that correctly?

• 0 views • link

written 1.8 years ago by avril •