GoKawiil
Tech News
A group of drug companies trained an open-source AlphaFold 3 replica called OpenFold3 on more than 20,000 proprietary protein structures pooled from their private vaults. The resulting model outperformed versions trained only on public data or on any single firm's siloed dataset, according to an unpublished, non-peer-reviewed blog post.
Apple has published a new study introducing SimpleDesign, an AI model that generates protein sequences and their 3D structures together in one process rather than sequentially. It builds on Apple's earlier SimpleFold work, which used flow matching paired with general-purpose Transformer blocks to predict protein structures from amino acid sequences more efficiently than models like AlphaFold.
A research team mined all 214.5 million structure predictions in Google DeepMind's AlphaFold database to search for undiscovered 7TM (seven-transmembrane) folds resembling G-protein-coupled receptors. Using the rhodopsin structure as a reference and the TM-align algorithm, they identified TM184C as a previously unrecognized GPCR-like protein involved in regulating cell-to-cell exchange and autophagy. The analysis required processing 24 tebibytes of compressed structural data on university supercomputing clusters.
Researchers solved eight cryo-EM structures of human UGCG, the enzyme that gatekeeps glycosphingolipid synthesis, capturing it alone and bound to substrates, products and the drugs miglustat, ibiglustat and eliglustat. The team also ran molecular dynamics simulations and a cross-species sequence comparison spanning platypus, marsupials, elephants and other mammals to trace how UGCG regulation evolved, finding features unique to primates. All structural data, simulation files and sequences have been deposited in public repositories.