Remember Golden Gate Claude? Etowah Adams and I have been working on applying the same mechanistic interpretability techniques to protein language models. We found lots of features and they’re... pretty weird? 🧵

Brilliant video by Veritasium on the development of protein structure prediction featuring #AF2 and #Rosetta series with David Baker youtu.be/P_fHJIYENdI?si…

Are protein language models the new universal key? doi.org/10.1016/j.sbi.… Brilliant and thoughtful piece.

Lab-in-the-loop therapeutic antibody design At Prescient Design Genentech we have spent 3+ years reimagining drug discovery. We built a machine learning system to design and execute experiments. Here's how it works and what we can do 🧵 1/

Half of an AI scientist is rejecting or accepting hypotheses. FutureHouse and ScienceMachine just put out ~300 novel hypotheses from ~50 published papers along with ground-truth data. Humans take 4.2 hours to solve these and frontier models get 10-20% correct. SWE-bench for bio

Predicting the conformational flexibility of antibody and T-cell receptor CDRs - This paper presents ITsFlexible, a novel graph neural network (GNN) model designed to predict the conformational flexibility of Complementarity-Determining Region (CDR) loops in antibodies and

Protein function often depends on protein dynamics. To design proteins that function like natural ones, how do we predict their dynamics? Hannah Wayment-Steele and I are thrilled to share the first big, experimental datasets on protein dynamics and our new model: Dyna-1! 🧵

CONSTRUCT: an algorithmic tool for identifying functional or structurally important regions in protein tertiary structure 1. CONSTRUCT is a new computational tool designed to identify clusters of conserved amino acid sites in protein tertiary structures—regions likely critical

The Steinegger lab found a single putative novel fold among 821M predictions. Once again, this plot comes to mind: PLMs are only as good as the number of homologs in the training set Metagenomic-scale analysis of the predicted protein structure universe biorxiv.org/content/10.110…

CreoPep: A Universal Deep Learning Framework for Target-Specific Peptide Design and Optimization 1. CreoPep is a generative deep learning platform that enables target-specific peptide design by combining masked language modeling with progressive masking and energy-based

🧵5 Top Free Alternatives to BioRender for Scientific Illustrations! These five websites offer free scientific illustrations for biologists. Great for presentations, research papers and other research communication needs. Save and share the post!

AI-Guided Discovery and Optimization of Antimicrobial Peptides Through Species-Aware Language Model １．This study introduces LLAMP, a species-aware AI model for predicting antimicrobial peptide (AMP) activity. LLAMP significantly outperforms existing models in predicting MIC

Ever wondered how to spot new antibiotics hiding in biological data? Our latest paper in Nature Protocols Nature Portfolio offers a practical, end-to-end ML playbook for mining genomes & proteomes—AI vs AMR, one dataset at a time. Link: rdcu.be/el2Go

Our latest work is out: we designed dual GLP1R/GCGR agonists—cyclic peptides that activate both metabolic receptors, entirely from sequence alone. This has never been done before. 🔗 biorxiv.org/content/10.110…

1/2 🧵| 2 MUST read papers if you want to use generative AI with proteins. tldr: Diff create + plausible but - diverse proteins, PLMs do the opposite biorxiv.org/content/10.110… Among diff models, RFDiffusion & Chroma exhibit the most balanced performance arxiv.org/abs/2504.16479

I'M BLOWN AWAY. Andrej Karpathy just explained Software 3.0 at YC. BIG IDEAS: English is coding. AI is electricity. And, build for LLMs, not just people. Key takeaways:

📢 Call for proposals: Boltz small-molecule design collaboration! 🧬 Can we help design your ideal molecule? Can you help us improve our open-source models? Please reach out or share with scientists you know! More details below! It has been great to see the level of excitement

Is Chain-of-Thought Reasoning of LLMs a Mirage? ... Our results reveal that CoT reasoning is a brittle mirage that vanishes when it is pushed beyond training distributions. This work offers a deeper understanding of why and when CoT reasoning fails, emphasizing the ongoing

Happy to announce that PepMLM is now published in Nature Biotechnology Nature Biotechnology after 2.5 years (in the pre-RFDiffusion era 🥲) A brief recap: We simply concatenate the target protein sequence with the peptide binder sequence, mask the whole binder region, and fine-tune

I've written up some thoughts on publishing for machines. 10M research papers are published per year and there are 227M total - machines will be primary producers and readers of publications going forward. It's time to revise the scientific paper.