The great oxidation continues!

Deep QC should run on every sequencing dataset. In a new post, Fulcrum co-founder @tfenne.bsky.social explains why he built Riker, a modern successor to Picard designed to make rich sequencing QC fast enough to run every time.

Blog: shorturl.at/oUTXk
Repo: github.com/fulcrumgenom...

This is a big personal win for me. I and my team wrote Picard when I was at @broadinstitute.org and it's been used in so many pipelines. It's really satisfying to revisit it, speed it up, make it more ergonomic, and slim down the distribution.

github.com/fulcrumgenom...

GitHub - fulcrumgenomics/twistcgp: Nextflow pipeline for Twist Comprehensive Genomic Profiling (CGP) panel analysis Nextflow pipeline for Twist Comprehensive Genomic Profiling (CGP) panel analysis - fulcrumgenomics/twistcgp

Oncology pipelines don’t stay where they were first built. They move across teams, infra, and timelines, and can start to get brittle.

So we built twistcgp with @twistbioscience.com as an nf-core-based workflow for analyzing data from the Twist Oncology DNA CGP Panel.

github.com/fulcrumgenom...

Lots of solid points. For me the maintenance argument is strong - if you're not going to support it, don't port it!

Amen! Rust just got so much right in the building, packaging, publishing, and installation space. Every other language I touch could learn so much from it.

The correct link: lh3.github.io/2026/04/17/t...

I should add that I'm also trying to do this from low-pass WGS data in the 0.25x-2x range.

@brent-p.bsky.social I saw you were adding contamination detection to somalier - is there a writeup of the method(s)? I'm making good use of VerifyBamID2's SVD/PCA based approach to help me out with a population ("domestic" cats) that it seems nobody understands sufficiently well yet.

Spoiler: it's mostly caching/memoizing pow()/log() calls that are all on a small, fixed domain of values (quality scores).

Performance tuning of FREEMIX calculations from BAM by tfenne · Pull Request #82 · Griffan/VerifyBamID Hi @Griffan - I've been running VerifyBamID2 on some fairly large panels (because my goal is to get to very low coverage), and spent some time working on the runtime of the process for calculat...

But the one I'm really excited about is the one where Claude and I made the optimization phase of contamination prediction (everything after BAM pileup) about 20X faster: github.com/Griffan/Veri...

And to show it's not all rewrite-in-rust this week ... with Fan Zhang's kind help, I've had a number of PRs merged into VerifyBamID2 that allow it work seamlessly on non-human references, and fix some bugs I ran into along the way.

Features include: whole genome or region based sampling, insertion of variants from a "standard" VCF, generation of a "golden BAM" with correct alignments of all reads, and of course it's blazingly fast 😜

I did search, and couldn't find any other simulators that allowed for insertion of variants from a VCF, only over regions specified in a BED file, and was actively maintained 😱

GitHub - fg-labs/holodeck: NGS read simulator NGS read simulator. Contribute to fg-labs/holodeck development by creating an account on GitHub.

I was sitting in talks at AGBT Ag this week, trying to simulate reads over SNP sites to bootstrap my way into doing contamination detection in domestic cats, when I got pretty frustrated with DWGSIM by @nilshomer.com who happened to be sitting right next to me. Now I have github.com/fg-labs/holo...

At #AGBTAg26, @nilshomer.com will present our work with @broadinstitute.org , @umasschan.bsky.social, and Darwin's Ark on building a comprehensive imputation reference panel for domestic cats.

See the flash talk today at 4:30pm during the Plenary, or catch poster #128 during the poster sessions.

Your freezer is lying to you. 🥶 We found frozen samples picked up way more artifactual C>T mutations than DNA stored at room temp. The data surprised us. 🧬👇

Next step is to get this integrated into HTSJDK so that it can ripple out to Picard, GATK and other tools that rely on it for BAM, CRAM and other formats that use gzip/DEFLATE compression.

Benchmarking results on a 2.5G SAM file at AWS showing 3.75-4.75X faster compression and 2-2.7X faster decompression than the JDK's built in methods.

On a small AVX-512 enabled instance an Amazon it runs up to 5x faster than the JDK's zlib compression, and over 2.5x faster for decompression!

There's a 0.1.0 release up on github and in sonatype for anyone that would like to try it. The JAR contains native libs for windows, linux/x86_64, linux/aarch64, mac/x86_64, and mac/aarch64. And a small benchmarking tool...

GitHub - fulcrumgenomics/jlibdeflate: Java JNI wrapper around libdeflate for faster DEFLATE/gzip/zip on the JVM Java JNI wrapper around libdeflate for faster DEFLATE/gzip/zip on the JVM - fulcrumgenomics/jlibdeflate

It's amazing to me that libdeflate has been around for ~a decade and there hasn't been a production-ready JNI wrapper for it in all that time. So this week I built jlibdeflate: github.com/fulcrumgenom...

The great oxidation is happening!

Sure ... but those same people are so risk averse (or constrained) they likely won't update even if the outputs are identical at the byte level 😱

I mostly agree with what you're saying at rewrites.bio - though some tools in bioinformatics need more than a rewrite! When we can surpass old limitations, or modernize output formats a "new tool heavily inspired by" an existing tool can be a better choice than a rewrite with shared baggage!

@robp.bsky.social you were asking a few weeks ago for AI rust porting examples. Here's probably our biggest to-date. All the UMI tooling from fgbio. @nilshomer.com invested a lot of time into this beyond straight porting (e.g. massively better concurrency in many areas).

If you use our fgbio UMI tooling today, you owe it to yourself to check out fgumi! Massively faster re-write into Rust by @nilshomer.com with 100% reproduction of the fgbio outputs!

Ever receive a BAM aligned to “GRCh38” and realize that could mean several different references?

We just released ref-solver, a tool that identifies the reference genome behind BAM/CRAM/FASTA files by comparing sequence dictionaries to known builds.

🔗 whatsmygenome.fulcrumgenomics.com

Read more 👇

100%. The problem is that the ethical concerns require governmental or international solutions. Opting out of AI in a technical field right now is a good way to be instantly out competed by those willing to adopt!

Such a cool project from @nilshomer.com.

GitHub - fulcrumgenomics/refget-rs: GA4GH refget Sequences v2.0.0 and Sequence Collections v1.0.0 — server, client, and CLI tools in Rust GA4GH refget Sequences v2.0.0 and Sequence Collections v1.0.0 — server, client, and CLI tools in Rust - fulcrumgenomics/refget-rs

Went looking for a full-stack GA4GH refget implementation in Rust. Couldn't find one, so I had Claude build it. 🦀 refget-rs server, client, data models, storage, & CLI for Sequences v2.0.0 + Sequence Collections v1.0.0. Serve, query, compare, compute digests. 🔗 github.com/fulcrumgenomics/refget-rs