LSTMs strike back

Getting ready for what is sure to be an awesome panel on translation of ML into the life sciences with Michael Bronstein Christian Dallago Andrew Phillips and Sepp Hochreiter!

P(L)INDER

Our very own Kieran Didi NVIDIA NVIDIA Healthcare discusses the ups and downs of equivariance in life science and materials!

You don’t get a tweet from Michael Bronstein everyday. Definitely framing this one.

Evolutionary Multiobjective Molecule Optimization in an Implicit Chemical Space pubs.acs.org/doi/10.1021/ac… #JCIM Vol64 Issue13 #compchem

📸 Happening tomorrow - 11am ET / 5pm CET / 8am PST. Sign up here: genaiindrugdiscovery.com!

Protein language models learn the evolutionary rules of natural sequences, but for design we need methods to steer them. I'm excited to share some of our work on this from Profluent and introduce proseLM: protein structure-encoded language model. biorxiv.org/content/10.110…

🚀💯

I don't think I've been this excited about a geometric deep learning paper in a long time 🤯 Gibson, Tubbenhauer and Williamson: Equivariant neural networks and piecewise linear representation theory

1/ 🧬 Excited to share CHEAP, our new work on compressed protein embeddings. We characterize the joint distribution of p(sequence, structure) in ESMFold's latent space, and find cool tidbits on compressibility, tokenizability, and pathologies: biorxiv.org/content/10.110… 🧵

How do we get all-atom structure & sequence embeddings for biomolecules *from sequence alone?* In exciting new work led by Amy Lu, we introduce CHEAP - a general method using compression and quantization for joint embeddings! Explainer 🧵 and preprint below 👇

Automated molecular optimization of a BRD4 inhibitor using #AlphaFold predicted structure with our new #GenerativeAI drug design method, DrugHIVE. paper: pubs.acs.org/doi/10.1021/ac… GitHub: github.com/jssweller/Drug… Remo Rohs lab USC Dept. of Quantitative & Computational Biology. JCIM & JCTC Journals. #compchem #machinelearning

🔥 Don't forget to join us this Friday - 5p CET/11a ET! x.com/vant_ai/status…

Alphafold3 reproduction.

Please checkout out P(L)INDER training workshop and submit your entry for the challenge.

Join us on Tuesday! plinder.sh/blog/training

DPLM-2: A Multimodal Diffusion Protein Language Model • DPLM-2 introduces a novel framework for generating protein sequences and structures simultaneously, eliminating the need for separate models. • It leverages a token-based representation for 3D structures, enabling

Proud of our VantAI team for this amazing work.

Neo just got much faster!💥 Working closely with NVIDIA Healthcare, VantAI adopted #cuEquivariance, leading to massive speedups. Today's announcement is a big advancement for the field!