search

jmonlong / manu-vgsv

https://jmonlong.github.io/manu-vgsv/
Other
9 stars 2 forks source link

Mapping comparison vg vs hg38-alt #12

Open jmonlong opened 5 years ago

jmonlong commented 5 years ago

Minutes from yesterday's meeting:

@glennhickey is currently working with Peter on a mapping experiment comparing mapping to vg vs mapping to the linear reference augmented with alts.

We can use this issue to discuss results or where to put them in the text etc.

glennhickey commented 5 years ago

Preliminary results not great for this. I re-aligned AFR_BantuSETswana_HGDP01030_M from Peter's table S2S. He had 14372109 unmapped when aligning to grch38 w/alts, and 13514498 when adding in his sv alts. With vg, I got 15467529. I'm doing another sample now, but we may have to try restricting the analysis to reads that map to SVs or something in order to get any signal.

glennhickey commented 5 years ago

It took a while, but I got a SV graph built with (most of) the GRCH38 alt sequences built in via msga alignment. Remapping the same sample gave 13355882 unmapped reads. So we're aligning slightly more reads to the graph than Peter's alt contig based reference.

jmonlong commented 5 years ago

Did you have new numbers for this? I think you mentioned that decoy were affecting some of the results where vg looked better?

glennhickey commented 5 years ago

I had a good result where we were mapping more reads than bwa-mem on Audano's pan-genome. But then I realized that I was including the decoy sequences from HS38D1, whereas they were not. Dropping the decoys, we mapped more than the normalize reference but less than Audano.

I don't think counting mapped reads is super meaningful, as shown by the fact that the decoy sequences have much more impact on this than all the SVs combined (same for the GRCH38 alts which I had to align as contigs into the graph).

Something more precise where we count high MAPQ reads around SVs is needed. I think completing the current manuscript should take priority over this, but I'll check to see if I have scripts to pull this out of vg.

On Wed, Mar 13, 2019 at 3:12 AM Jean Monlong notifications@github.com wrote:

Did you have new numbers for this? I think you mentioned that decoy were affecting some of the results where vg looked better?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/jmonlong/manu-vgsv/issues/12#issuecomment-472306285, or mute the thread https://github.com/notifications/unsubscribe-auth/AA2_7s8S0goJ65mq0LCIMKOPXhrigK7Fks5vWKTQgaJpZM4akUIs .

jmonlong commented 5 years ago

I totally agree, thanks for the update.