Martin Steinegger 🇺🇦

Looking very much forward to it!

#RECOMB2026 will be in Thessaloniki, Greece on May 26-29, 2026. Satellites on May 24-25. Save the date!

Το συνέδριο #RECOMB2026 θα πραγματοποιηθεί στη Θεσσαλονίκη, στις 26-29 Μαΐου 2026. Οι δορυφορικές εκδηλώσεις θα διεξαχθούν στις 24-25 Μαΐου 2026. Σημειώστε την ημερομηνία!

Thank you Caroline! :)

We are excited to share GPN-Star, a cost-effective, biologically grounded genomic language modeling framework that achieves state-of-the-art performance across a wide range of variant effect prediction tasks relevant to human genetics.
www.biorxiv.org/content/10.1...
(1/n)

Exon finding seems very well suited for GPU acceleration. Worth revisiting exonerate. :)
DP remains very powerful and aligns well with AI approaches, whether via scoring schemes or tokenized data (e.g. Foldseek).

Thank you! Yes, it uses DP to compute the maximal ungapped score, followed by a GPU-based Gotoh–Smith–Waterman, so no k-mer index is required. The drawback is that you can’t trade sensitivity for speed, but full DP searches against UniProt in milliseconds open up many exciting applications.

Technically yes, but UniProt is highly redundant, so searches against an unclustered database could produce extremely long lists, potentially overwhelming the interface. What's your use-case?

One of the shared first authors just joined Bsky. Welcome Alex @achancond.bsky.social

This work was only possible through the great work of Felix Kallenborn, Alejandro Chacon, Christian Hundt, Hassan Sirelkhatim, @kdidi.bsky.social, @sooyoung-cha.bsky.social, @machine.learning.bio, @milot.bsky.social, Bertil Schmidt n/n

We are currently integrating Grace and Blackwell optimizations and further speeding up the algorithms in MMseqs2-GPU and structure prediction. Below is a sneak peak of our current progress. 5/n
📄 research.nvidia.com/labs/dbr/ass...

My first email to Johannes Söding, my later PhD advisor, proposed a GPU-accelerated HHblits. But GPUs in 2012 had many limitations. Now they are widely deployed and massive number crunchers. I am happy that together with @unimainz.bsky.social and NVIDIA we were finally able to build MMseqs2-GPU. 4/n

Homology retrieval grounds ML systems to produce reliable predictions. MMseqs2 is already used in Boltz1/2, BioEmu, MSA-Pairformer, Chai-1, BioNeMo, Proteinx, etc. MMseqs2-GPU can enable these and next-gen models to integrate fast homology retrieval for end-to-end GPU inference. 3/n

Below we show GPU-accelerated Foldseek, searching 128 structures against AFDB50 (54 million structures). On 128 CPU cores this takes ~120 seconds, whereas a single GPU completes it in ~25 seconds. 2/n

GPU-accelerated homology search with MMseqs2 - Nature Methods Graphics processing unit-accelerated MMseqs2 offers tremendous speedups for homology retrieval from metagenomic databases, query-centered multiple sequence alignment generation for structure predictio...

MMseqs2-GPU sets new standards in single query search speed, allows near instant search of big databases, scales to multiple GPUs and is fast beyond VRAM. It enables ColabFold MSA generation in seconds and sub-second Foldseek search against AFDB50. 1/n
📄 www.nature.com/articles/s41...
💿 mmseqs.com

GPU-accelerated homology search with MMseqs2 - Nature Methods Graphics processing unit-accelerated MMseqs2 offers tremendous speedups for homology retrieval from metagenomic databases, query-centered multiple sequence alignment generation for structure predictio...

GPU-accelerated MMseqs2 offers tremendous speedup for homology retrieval, protein structure prediction with ColabFold, and protein structure search with Foldseek. @martinsteinegger.bsky.social @milot.bsky.social @machine.learning.bio

www.nature.com/articles/s41...

EcoFoldDB: Protein Structure-Guided Functional Profiling of Ecologically Relevant Microbial Traits at the Metagenome Scale enviromicro-journals.onlinelibrary.wiley.com/doi/10.1111/...

YouTube video by ProteinDesignStudio Screen binders using ipSAE

pip install ipsae
from www.linkedin.com/in/ullah-sam...

www.youtube.com/watch?v=A5ph...
PyPI pypi.org/project/ipsae/
His github fork github.com/ullahsamee/I...
My github github.com/DunbrackLab/...
Paper www.biorxiv.org/content/10.1...
For designed protein binders www.biorxiv.org/content/10.1...

Preprint:
Highly efficient protein structure prediction on NVIDIA RTX Blackwell and Grace-Hopper
nvda.ws/4n4xzz9

Visit the NVIDIA Digital Biology Labs website to find more information like this:
t.co/R9ufEZrGEA

De novo discovery of conserved gene clusters in microbial genomes with Spacedust - Nature Methods This work presents Spacedust, a tool for de novo identification of conserved gene clusters from metagenomic data.

Spacedust: a tool for de novo identification of conserved gene clusters from metagenomic data.
@ruoshiz.bsky.social @milot.bsky.social

www.nature.com/articles/s41...

hey bluesky 👋 visa hurdles mean I’m looking for opportunities outside the US. I’m a computational biologist (bacterial + phage genomics, postdoc in Koonin’s group @ NIH). I am interested in teaming up on funding apps. reach out if this resonates!

Was about to write the same.

Funny, I had the same question on my mind today.

@pedrobeltrao.bsky.social here you can see the clustering at 90% identity.

Also, this dataset only contains complete ORFs (containing both a start and stop codon). Metagenomic samples, especially Soil are often very harder to assemble, which frequently results in incomplete ORFs.

🌎👩‍🔬 For 15+ years biology has accumulated petabytes (million gigabytes) of🧬DNA sequencing data🧬 from the far reaches of our planet.🦠🍄🌵

Logan now democratizes efficient access to the world’s most comprehensive genetics dataset. Free and open.

doi.org/10.1101/2024...

Exciting to see our protein binder design pipeline BindCraft published in its final form in @Nature ! This has been an amazing collaborative effort with Lennart, Christian, @sokrypton.org, Bruno and many other amazing lab members and collaborators.

www.nature.com/articles/s41...

Does anyone know of a recent comparison of the main structural classification schemes of proteins and guidance on when to choose one? Something like this but including ECOD and perhaps seq-based schemes like Pfam, SUPERFAMILY and CDD.

Img source (2020)
pubmed.ncbi.nlm.nih.gov/32302382/

#structuralphylogenetics #strphy #3di

Congratulations!

The ColabFold server has 80 cores (Intel Xeon E7-8891 v2) with 4 TB RAM (using ~1 TB).