flâneur

Portal

portal.valencelabs.com · 431 words · saved by 1 readers

I previously summarized some debate around a method called AF-Cluster (see the post here: https://portal.valencelabs.com/ml-for-proteins/post/active-scientific-debate-on-generating-conformations-with-alphafold-KbrktEWEf7S60TC) Now, Wayment-Steele and co-authors respond to concerns about reproducibility with a new preprint, arguing that the critiques used older AF2 params that weren't directly comparable with what they were using. They even provide a new colab notebook to help the field compare and combine sampling methods. This seems like a nice response to reproducibility to me; trying to be transparent and developing resources to help bridge divides. https://www.biorxiv.org/content/10.1101/2024.07.29.605333 There have been a few papers recently addressing the huge resources needed for training protein language models (PLMs) and proposing solutions. Nathan Frey and a group at Genentech discussed "Cramming Protein Language Model Training in 24 GPU Hours", training a model in a single d

Inside Valence A behind-the-scenes look at the research at Valence Labs, exploring new ways to predict, explain, and ultimately decode biology. Over 2,000 subscribers By subscribing, you agree Substack's Terms of Use, and acknowledge its Information Collection Notice and Privacy Policy. Home Archive About The Virtual Cell Comeback An introduction to simulating biology READ THE LATEST Valence Blog View all Truly All-Atom Boltzmann sampling with AquaGen We built a model that shows it’s possible to train a Boltzmann sampler that still works with a key ingredient of biochemical…

saved by

related reading