https://nf-co.re logo
Join Slack
Powered by
Just curious but what is the status of adding host...
# taxprofiler
c
Just curious but what is the status of adding hostile or other host removal tools? We’re using the current built in method using a non-model database (Gallus gallus) but we are seeing bleed through of host reads which suggests bowtie2 parameters may need some tweaking (reads seem to be from repetitive regions)
j
Uhhhh for hostile looks like I got 95% of the way and then we forgot 😅
I could try porting it over next week... but I don't know how long that would take
Or I try now for funsies...
c
No worries, I wasn’t sure. I thought about forking and implementing, but the impression I had from the pull request you set up was to make this a bit more pluggable
j
Yeah
So I got 95 % of the way making the updated PR this morning and got to the same point as before which was testing 😅
(can't test the default method because the prebuilt reference genomes are too large/require too much memory)
So I'll see if I can come up with a work around (I have an idea)
c
No hurry on it, I’m testing a workaround, basically adjusting the bowtie2 parameters to be a bit more relaxed on alignments. Saw a jump from ~40% host to ~75% host-aligned reads using 
--local --dovetail
, I suspect 
--local
is the major impact since it soft-clips.
Question is whether this is over-filtering now, but the majority of those were proper pairs so I think it’s a little closer to the truth in our samples
Just thinking, but looking at what hostile is doing under the hood: taxprofiler has essentially reimplemented it in the workflow already, correct?
👀 1
taxprofiler is using bowtie2 for short reads, minimap2 for long reads. Is the magic sauce in hostile the database?
j
Yes, the database is the biggest thing. But there are some other nice convenience things that also makes it nicer (all in one tool, nice reporting etc)
🙏 1
The new kid of the block is @Bede Constantinides new tool 'deacon' which is what we will add next
👍 1
b
Cool!
j
But in the meantime @Bede Constantinides I made a feature request on hostile
(and same applies to deacon )
😬
If you have a spare 20m (which inevitably will be like 8h because bioinformatics translation of time)
😄 1
s
Just thinking, but looking at what hostile is doing under the hood: taxprofiler has essentially reimplemented it in the workflow already, correct?
It does look a lot like that 🙂
🥲 1
j
But I don't want to maintain it ;)
💯 1
b
Hey James, yes this seems like a good idea, let me add some small indices for testing
❤️ 1
j
When Bede does it for me 😬
c
Yeah it’s been mentioned elsewhere, host removal is particularly difficult. We work w/ essentially everything non-human/mouse, so we’re limited by the quality of the host genome(s).
Pangenomes will be the way to go eventually (also discussed here) 😉
j
Yes, thus deacon (which I'm very excited about?
❤️ 1
💯 3
)*
b
Pangenomes will be the way to go eventually (also discussed here)
They already are!
panhuman-2
will be landing soon with better host classification sensitivity and specificity
c
Well, in our case we need non-human 🙂
b
Which taxa?
c
Gallus
b
We can build one for chicken – do you have a collection of genomes in mind?
Building 
panhuman-1
from 96 genomes takes about an hour on my laptop
c
FASTA or from variant graph?
b
Currently it just accepts linear genomes
c
That’s probably for the best right now 🙂
b
But that doesn’t make a practical difference does it?
Not for k-mer decomposition anyway
c
Not really, it’s mainly working with a new format
b
Ok well feel free to open a feature request if you think it would be a useful feature in deacon
I’ve tried to make it as easy and quick as possible for people to build (and/or mask) their own pangenome indexes
c
I’ll definitely point deacon out to our collabs, they’re working on African data sets and want to screen samples using their own assemblies + HPRC release 2
👍 1
I’m very tempted to test it on NCBI Gallus assemblies and compare to what we have. The main task now is which ones to select!
👀 1
🐔 1
b
@James Fellows Yates As requested in issue https://github.com/bede/hostile/issues/70 I’ve added a new index 
human-mit
containing just the human mitochondrion to both Hostile and Deacon S3 buckets (hosted by Uni of Oxford). For Hostile use 
hostile index fetch -n human-mit
and for Deacon use 
deacon index fetch human-mit
j
@Bede Constantinides small nudge on this bug report: https://github.com/bede/hostile/issues/71
It's currently preventing us getting 
hostile
officially into taxprofiler
(if you're back from holiday/travelling)
✅ 1
b
Oh I’m sorry about that, I’ll fix asap
j
Also really nice podcast with the µbinfie people 🙂
Enjoyed it a lot, you're very clear with your explanaitonS!
❤️ 1
b
Thanks! Was fun. Wish I knew a bit more about taxprofiler at the time!
Thanks, my fear with such things is rambling incoherently
@James Fellows Yates should now be sorted
❤️ 1
j
Ok, testing now!
Thanks, my fear with such things is rambling incoherently
Nah, your fears are unfounded !
2 Views
Open in Slack
PreviousNext
Powered by