AlphaFold2 seems to be: - 3D equivariant Transformer which "builds physical & geometric notions into the architecture": fabianfuchsml.github.io/alphafold2/ (Fabian Fuchs) - Iteratively predict 'distograms' which are used to update the underlying message-passing graph for the Transformer.

Because deep learning is so empirical, success in it is to a large extent proportional to raw experimental throughput - the ability to babysit a large number of experiments at once, staring at plots and tweaking/re-launching what works. This is necessary, but not sufficient.

Physics computation itself acts as a regularizer while training the model embedded in it. Our results serve as proof-of-principle to smear the boundary between physics computation and ML. We are glad to share our work with you on Physical Review Letters #openaccess journals.aps.org/prl/abstract/1…

Inverse design of 3d molecular structures with conditional generative neural networks. (arXiv:2109.04824v1 [cs.LG]) ift.tt/3k4OQdH

Many thanks to our members John Vant Eric Wilson John Kevin Cava Nicholas Ho Chitrak Gupta @17Dsarkar Mrinal Shekhar Sumit @estreitwieser Fathima Doole Calvin Chun Kit Chan, Ph.D who work 24x7 to build a Research program !! Onto mid-tenure review with ASU School of Molecular Sciences :)

Have you wondered why object detection, unlike classification, has so many sophisticated algorithms? With Pix2Seq (arxiv.org/abs/2109.10852), we simply cast object detection as a language modeling task conditioned on pixels! (with Saurabh Saxena, Lala Li, David Fleet, Geoffrey Hinton)

Sequence vs Graph, which representation is better? Obviously GNNs can predict per-node, while sequences can't. But what about per-graph properties? We can now bijectively encode graphs as sequences, so it's past time that we answer this for AI in chemistry.

End-to-end learning of multiple sequence alignments with differentiable Smith-Waterman biorxiv.org/content/10.110… A fun collaboration with Samantha Petti, Nicholas Bhattacharya, Roshan Rao, Justas Dauparas, Neil Thomas, JuannanZhou, Sasha Rush & Peter Koo (1/8)

Thanksgiving update. #ML4Molecules workshop paper NeurIPS Conference By John Kevin Cava John Vant Ankita Shukla Abhishek Singharoy Nicholas Ho, and Ross Maciejewski. “Towards Conditional Generation of Minimal Action Potential Pathways for Molecular Dynamics". Happy thanksgiving !!

Learn more about Meta AI’s newest inverse protein folding model: ESM-IF1, developed by Meta researchers and team Chloe Hsu Adam Lerer Brian Hie Tom Sercu Robert Verkuil Jason Liu and Zeming Lin Check it out: 👇

Years after being told my interests don't line up with planning, I literally can't believe I am doing planning for Molecular Dynamics. 🤣

The MLSB 2022 Call for Papers is up on our website! This year, we're looking for 4-5 page short papers. Submission deadline is Thursday, September 22nd, 2022 AoE. Check out the website (mlsb.io/#callforpapers) for more info and come to New Orleans in December!

Learning Free Energy Pathways through Reinforcement Learning of Adaptive Steered Molecular Dynamics biorxiv.org/content/10.110…

Happy to announce that our paper on learning free energy pathways has been accepted to MLSB at NeurIPS 2022! MLSB @ NeurIPS

Procrastinated a bit much on making this tweet, but I'm happy to say that our paper on using RL + MD to learn pathways was accepted to MLSB (in San Diego + Copenhagen)! Happy to talk about it, what worked, what didn't, and the next steps!

Extremely excited to be in person at #NeurIPS2022 . I will be here until Dec 3rd. Anyone interested in SciML, applications to Molecular Biology, DM me and lets have a chat!

Use a VAE to predict atomic coordinates from coarse-grained coordinates (backmapping) for any protein system. Soojung Yang RGB Lab @ MIT arxiv.org/abs/2303.01569

Happy to present "Latent Sequence Generation of Steered Molecular Dynamics" as a poster in the #Physics4ML workshop at #ICLR2023 openreview.net/pdf?id=iPJazFv…

To internship or not to internship next summer? 🤔

Out in Cell Systems, Method to find peptide binders from human proteome in 7 seconds ! Eric Wilson's best work :) 🙏Diego Chowell Access: hlainception.asu.edu:3000 Release: eurekalert.org/news-releases/… Paper: authors.elsevier.com/a/1ir1M8YyDfql… Biodesign Institute at Arizona State University Mayo Clinic Mount Sinai Health System