Kyle Tretina

@allthingsapx.bsky.social on the new Biohub initiative by the Zucks

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

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

Near-real-time protein structures change science:

It means:
→ Next-gen protein AI data waves
→ Interactive protein design loops (DMTA in hours)
→ Proteome-scale insights with fewer resources

It means the bottleneck doesn't have to be compute.

It's close (preprint below).

+1

Does anyone here care about biomolecular AI? Who should I follow?

It looks like each per‑residue latent captures almost exclusively local information

When the authors perturb the latent of a single residue, only that residue’s reconstruction quality changes, while others stay intact

Paper: arxiv.org/html/2507.09...
Project: t.co/TuplU054vw

🧬 Introducing La‑Proteina:

a partially‑latent flow‑matching model that co‑generates sequence + all‑atom structure for proteins up to 800 aa 🧬

Side‑chains live in latents, backbone explicit → 75 % codesign & SOTA motif scaffolds 🔥

I'm at ICML 2025!

DM me if you want to chat.

@icmlconf.bsky.social #ICML2025 #icml25 #BioNeMo

Accelerated Sequence Alignment for Protein Science with MMseqs2-GPU and NVIDIA NIM | NVIDIA Technical Blog Protein sequence alignment—comparing protein sequences for similarities—is fundamental to modern biology and medicine. It illuminates gene functions by reconstructing evolutionary relationships…

To learn more about the impact of MMseqs2-GPU and test it out (for free), here's a blog I wrote:
developer.nvidia.com/blog/acceler...

Boltz-2 just dropped: open-source AI that predicts both protein complex folds ✚ binding affinities in one shot 🚀

This is a win for protein AI, but let's not forget MSAs, the bioinformatics backbone many structure models lean on.

msa-search Model by Colabfold | NVIDIA NIM Generates a multiple sequence alignment from a query sequence and a protein sequence database search.

Try it out:
build.nvidia.com/colabfold/ms...

Blog:
developer.nvidia.com/blog/acceler...

MMseqs2-GPU is available as a downloadable NVIDIA NIM microservice (MSA-Search)!

📢📢 "Proteina: Scaling Flow-based Protein Structure Generative Models"

#ICLR2025 (Oral Presentation)

🔥 Project page: research.nvidia.com/labs/genair/...
📜 Paper: arxiv.org/abs/2503.00710
🛠️ Code and weights: github.com/NVIDIA-Digit...

🧵Details in thread...

(1/n)

Kyle Tretina, Ph.D. ‪Insilico Medicine‬ - ‪‪Cited by 802‬‬ - ‪Molecular Microbiology and Immunology‬

Please add me 😄

scholar.google.com/citations?us...

I’m @neurips24! Let’s chat😁

"Are there at least 3 simulations per simulation condition with statistical analysis?"

From @commsbio.bsky.social's "Reliability and reproducibility checklist for molecular dynamics simulations" (doi.org/10.1038/s420...)

IMO the number 3 is meaningless and could equally well be 1 or 1000

Super excited to preprint our work on developing a Biomolecular Emulator (BioEmu): Scalable emulation of protein equilibrium ensembles with generative deep learning from @msftresearch.bsky.social ch AI for Science.

www.biorxiv.org/content/10.1...

Deep-Learning Based Docking Methods: Fair Comparisons to Conventional Docking Workflows The diffusion learning method, DiffDock, for docking small-molecule ligands into protein binding sites was recently introduced. Results included comparisons to more conventional docking approaches, wi...

DiffDock was the first time a traditional drug discovery simulation task was represented as a generative AI task AFAIK.

Recent DiffDock versions + other DL models are advancing rapidly + solving real problems for researchers.

Let's have a balanced conversation about it.
arxiv.org/abs/2412.02889

#CASP16 results are in! Template-based VFold seems to be lead method for nucleic acid structure prediction! AlphaFold2 and 3 still seem to be best methods for protein monomer and complex prediction.

Please add me to this starter pack. I'd appreciate it!

Please add me to this starter pack. I'd appreciate it.

Please add me to this starter pack. I'd appreciate it.

Please add me to this starter pack. I'd appreciate it.

Please add me to your starter packs. I'd appreciate it. Nice to meet you!

Please add me to this starter pack. I'd appreciate it.

Please add me to this starter pack. I'd appreciate it.

Please add me to this starter pack. I'd appreciate it.

Please add me to this starter pack. I'd appreciate it.