Had a great time at the #FASEB conference on illuminating the druggable proteome. Nice to catch up with old colleagues and sharing our research on GraphML for biology.

Check out the great work from our amazing intern Yijun Tian

AutoGluon 1.0 is live!! Shatters SOTA, wins 75% vs prior release, 63% win-rate vs best-in-hindsight combination of other methods. To our knowledge, this is the biggest leap forward in tabular ML in the past 4 years. See how we did it: github.com/autogluon/auto… #AutoML #AutoGluon

We at AWS AI Research & Education (AIRE) are hiring PhD students as Applied Scientist Interns for Summer '24 to work on AI for Science research projects related to biology, chemistry, and simulations. DM me for more info or apply @ amazon.jobs/en/jobs/242688…

Such a refreshing read for building virtual cell: arxiv.org/abs/2409.11654

LC-PLM: Long-Context Protein Language Model • LC-PLM introduces a structured state-space model (BiMamba-S) for protein sequences, enabling efficient modeling of long-range dependencies within protein structures, a capability that conventional transformer-based models struggle

📢 We're excited to introduce LC-PLM, a Long-Context Protein Language Model that redefines the potential of protein sequence modeling! LC-PLM addresses the challenges of modeling longer protein sequences and complex biological interactions. Our model demonstrates superior scaling

Check out our recent work for training protein language models with long-context capabilities!

strong results for mamba on protein modeling! firm believer in its efficacy on higher-resolution, long context data

A long-context protein language model with a bidirectional Mamba architecture. Yingheng Wang Zichen Wang, Gil Sadeh, Luca Zancato, Alessandro Achille, George Karypis Huzefa Rangwala

Excited to announce that we just released the LC-PLM model weights and related codes on GitHub: github.com/amazon-science…

Protein Structure Tokenization: Benchmarking and New Recipe 1/ Recent advances in protein structure tokenization (PST) methods enable direct application of language modeling techniques to protein 3D structures. However, the capabilities and limitations of these methods remain

Check out our recent work led by our outstanding intern Xinyu: we created the first benchmark specifically for protein structure tokenization and presents AminoAseed, a novel approach to address the codebook collapse problem in vector quantization.