Burstein lab

Fellow microbial immunity, mobile elements, and plasmid enthusiasts: the paper of our wonderful @BruriaSamuel on anti-defense genes in plasmids is now published in @Nature!!! Check out her thread below for highlights of our new results. 1/2

Very excited to see this work out today @ScienceMagazine! Discovering viral proteins that block immune signaling from predicted protein structures🤩 science.org/doi/10.1126/sc… Huge thanks to the amazing collaborators! 🤗 Linking previous thread on our findings below 👇

Out now! In collaboration with @LeifuChangLab, we uncover the molecular and structural underpinnings of CRISPR-Cas12f-like RNA-guided transcription systems! Links to the articles in the following tweet:

Evo 2, our genome language model that generalizes: - across biological prediction and design tasks, - across all modalities of the central dogma, - across molecular to genome scale, and - across all domains of life, is published today in @Nature.

Important context - Donald Knuth, one of the founding fathers of CS, is 88 years old

In new work in @jbloom_lab , we asked: how does endemic coronavirus keep evolving to erode human antibodies without disrupting spike function? biorxiv.org/content/10.648…

We’ve identified industrial-scale distillation attacks on our models by DeepSeek, Moonshot AI, and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models.

AlphaFast: High-throughput AlphaFold 3 via GPU-accelerated MSA construction 1. AlphaFast achieves a remarkable 22.8× speedup in end-to-end inference on a single GPU and up to 71.2× acceleration with four GPUs, reducing per-target runtime from nearly 15 minutes to just 8 seconds. 2. The core innovation replaces CPU-bound JackHMMER with GPU-accelerated MMseqs2 for multiple sequence alignment (MSA) construction, which was previously the dominant bottleneck accounting for over 95% of total runtime. 3. AlphaFast introduces a batched query architecture that consolidates unique sequences into a single GPU search, coupled with concurrent MSA post-processing and strict two-stage separation to resolve VRAM conflicts with JAX initialization. 4. Despite the massive speed improvement, AlphaFast maintains statistically indistinguishable structural accuracy from AlphaFold 3, with equivalent TM-scores, RMSD, pLDDT, and pTM metrics validated through bioequivalence testing. 5. The system supports both single-GPU and multi-GPU deployments with near-linear scaling, and includes a serverless implementation enabling cost-effective structure prediction at approximately $0.035 per target. 6. AlphaFast preserves the original AlphaFold 3 folding module, weights, and end-to-end performance, serving as a drop-in framework that decouples feature generation from inference for extensibility to other architectures. 💻Code: github.com/RomeroLab/alph… 📜Paper: biorxiv.org/content/10.648… #AlphaFold3 #ProteinFolding #StructuralBiology #GPUComputing #Bioinformatics #ComputationalBiology #MachineLearning #HighThroughput #ProteinDesign

🚀 Our results highlight a promising direction for improving the efficiency and scalability of protein language models. Read all about it! (5/5) biorxiv.org/content/10.648…

⚡ Reduced alphabets yield shorter inputs and major runtime gains, while maintaining comparable, and sometimes improved, predictive performance. (4/5)

🧬 New preprint alert! Protein language models have transformed bioinformatics, but what about the tokens they read? In our new preprint, 👑@EllaRannon👑 studies how tokenization choices shape pLM performance and efficiency. 🧵(1/5) biorxiv.org/content/10.648…

"All #T6SS-inducing regulators are equal, but some regulators are more equal than others". Preprint 🚨: We use #Vibrio to show that T6SS activation by regulator manipulation may result in the expression of different effector repertoires, affecting toxicity biorxiv.org/content/10.648…

Abstract submission is open for the SISB2026 meeting, aka Phage Defense! sisb2026.rockefeller.edu/abstract-submi… The deadline is March 16.

New paper out in PNAS!!! 🎉 Do more plasmid copies mean faster evolution? 🧵 Dive into the story pnas.org/doi/10.1073/pn…

🧬 THREAD: How bacteria evolved thousands of “smart syringes” to target specific cells Bacteria don’t just secrete toxins — some fire them with precision. Meet extracellular contractile injection systems (eCISs): phage-derived nanomachines that inject proteins directly into

Save the date! The CRISPR 2026 meeting will be held June 10 - 12 at Yale University, New Haven, Connecticut, USA. Abstract submissions will be opened on February 1st. Spread the word.

New preprint alert + a call for a community effort! 📢 🧬 ​Predicting host phenotypes (healthy/sick) from metagenomics is key to clinical translation. But while new tools keep coming out, researchers simply use Random Forest. Are newer methods worth the additional effort?

4/4 To allow standardized benchmarking for future tool development, we also released B-PPI-DB, a curated bacterial PPI database derived from STRING: doi.org/10.5281/zenodo… We hope B-PPI is just the first of many efficient bacterial PPI predictors!

3/4 B-PPI outperforms other rapid methods for bacterial PPI prediction without the high cost of structural folding and generalizes to unseen interactions with minimal fine-tuning.

1/4 Ever wanted to predict bacterial protein-protein interactions (PPI) on a large scale? We wanted to, but realized there’s no such algorithm that is both rapid and optimized for bacterial protein analysis. This led our ⭐️Chen Agassy⭐️ to develop B-PPI: doi.org/10.64898/2025.…