https://nf-co.re logo
Join Slack
Powered by
<@UPC8CHDKQ> &gt; The planned replacement fails va...
# sarek_dev
a
@Rike Hanssen
The planned replacement fails validation: GATK’s 
--output-cram-version 3.0
still writes CRAM 3.1 codecs internally, and Strelka rejects the file.
I tested a working alternative: put Samtools conversion inside ApplyBQSR’s existing task using Sarek’s combined image. Strelka passes, and all 5,642 decoded records match. This removes the extra process but retains conversion work.
meow tableflip 1
Also, after trying to add it to Strelka specifically:
Manta fails on the same direct GATK CRAM with `Failure to decode slice`; the indexed BAM control passes. A Strelka-only change would therefore leave Manta broken.
It could be an excuse to drop variant callers which don't support CRAM3.1 codecs 👀
m
SAME ISSUE
That is a pain
format label says 3.0 but it really is htsjdk 4.1+/5 So fails with strelka, and even fails with nft-bam 0.6.1 htsjdk 4
shakefist
m
the latest strelka release was 8 years ago... so you could say the software is “orphaned” if you catch my drift 😉
😆 2
👀 1
recompile with the new htsJDK and let claude fix any compat issues?
m
rust rewrite + update?
🤣 1
m
I’d start with recompiling 😄 not sure if anyone has the tokens to to a full rewrite
unless @Phil Ewels gets bored again of course
m
What do you think of Belka for the name?
m
Why Belca?
m
other soviet space dog in space that was paired with Strelka
that's after Laika
m
sure, sounds good to me 👍🏻
m
m
And it’s C++ 😅 I thought it was a java tool
m
meh, I can't read Java anyway
So C, python or whatever, is not an issue
😆
m
it’s even better I guess, maybe we could even just trigger a new conda build with the latest HTSlib
m
Yeah, let's try something simple before going wild
s
Strelka and Manta are both orphaned
(and have various unresolved issues)
m
Who has credits to burn
m
Ask fulcrum 🤣
🤣 1
f
But strelka is really widely used still so users would kill us if we just remove it haha
evil laugh 1
a
Us goading Phil into a Rust rewrite:
m
https://rea.tools/ Just saying
🤣 1
m
Seems like a prime target for the DRAGEN suite
m
BCLconvert?
m
And all the other stuff that’s integrated
like their GATK on steroids
m
Let's just recreate illumina
m
and all the other stuff they pulled away from github, including strelka
Let’s just recreate illumina
github.com/nf-illumina(ti)
f
Yes I want to recreate illumina
m
and instead of DRAGEN we call it TROLL
🧌 4
f
SMAUG
party this 1
party dead 1
m
also good 😄
the dragon that stole their hoard, seems fitting
m
Untitled
Untitled
f
and when we release directly ship it with sarek that would make headlines
100 rainbow 1
m
I kind of like 
Tenebra
m
We are live
m
Let us rise from the shadow of our corporate overlords 🙌🏻
🎉 1
Company needs a tagline 🤔
We steal your source code for the greater good
m
We liberate your code for the greater good
m
liberate does sound better
m
yeah, that's what I do with glasses at the pub, I liberate them
I don't steal
It sounds way better, right?
m
perception is everything
let’s profile it like a poor bioinformaticians Robin Hood
men in tights
a
In the interest of fairness, Dragen doesn't re-engineer OS tools, they're fresh implementations. They actually produce quite different results I think?
👍 1
but I wont stop you undercutting Illumina 👀
🎉 1
Oh FASTQC
m
As far as I know, most base tooling in DRAGEN was originally some open source project they archived and made closed source
shake no 1
m
What about FASTQC?
a
Dragen was always an Edico product before Illumina owned it
I believe the FPGA coding forces them to rewrite the algorithms? Not sure
👍 1
(Sentieon is the closed source version of GATK but faster)
f
well i think their best asset is anyways not the code or the tooling but the database they are now building by making dragen a product which is actually used
thistbh 2
they have all the samples
lets rewrite Sentieon aswell
talking about patient similarity search etc
m
I believe the FPGA coding forces them to rewrite the algorithms? Not sure
Yeah it does, but the base implementations came mostly from OSS I think
m
@Edmund is on board
Obviously
m
Lets rewrite Sentieon aswell
Decompiling that and “liberating” the speed gains would be cool
Pull the code and apply it to the OSS gatk
a
Easier to take GATK tools and rustify them, then guarantee parity
m
fulcrum is already on GATK
👍 1
m
Also fine 😄 Phiiiiiiiill ? Y’all got any more tokens?
m
They're trying to liberate code as well, let's not compete there
f
I have access to some locally hosted models (not the best of the best but infinite tokens)
m
Maybe Frontier can help out 😆
👌 2
@Famke Bäuerle, if you have GPU’s and free power I won’t stop you 😄
f
evil laughter
m
We have great responsibilities here, therefore, we have GREAT POWER
f
but not the time to set all of this up properly, so if someone does the backbone I can fire up some coding haha not sure how far we will get with the models we currently have (I think the best is deepseek v4.1)
m
@deepseek, rewrite gatk, don't make mistakes
👍 1
Should do the trick 🤷🏻 🤣
also source rewrites.bio for guidance
r
b3fmvn.jpg
party dead 2
🐕 3
☠️ 2
m
this is starting to feel like the javascript framework wars 😅
a
Oh we forgot about parabricks
m
f
but parabricks is free to use right?
m
Google is giving out free credits for their latest model
Copy code
For scientists & researchers: Apply for a chance to get access to Google Antigravity's highest usage tier free of charge for 6 months (onboarded on a rolling basis), subject to eligibility, capacity, and selection. Applications are now open.
a
but parabricks is free to use right?
free to use but the Nvidia GPUs aren't
m
Do you want to compete with NVIDIA and build GPUS?
I don't think we can vibe-code hardware
f
not yet
but if we have robots they can build the hardware and we vibe code their programming
robots everywhere
m
shared a GIF
Ok, so I have put in some more thoughts. I am so ready for this. Unless if we know about anyone already working on it
p
CleanShot 2026-10-16 at 18.35.56@2x.png
m
Yeah, but for Strelka/Manta
doit 1
p
Nice! Just remember that building the testing and validation rig is 10x the time/effort of rewriting the tools, and running the testing and validation is 100x
👍 1
Open in Slack
PreviousNext
Powered by