How accurately can one predict drug binding modes using AlphaFold models? New work from our lab reveals AF2's improved accuracy in capturing binding pocket structures, but the results of docking are not a slam dunk. ๐ฅ๐ง
Check out the preprint:
https://t.co/eS38wMoSoF
A new paper from our lab, @MIPS_Australia, and @KarunaPharma uses atomic-level simulations to reveal a molecular mechanism by which a ligand can achieve selectivity between nearly identical receptors. Open access:
https://t.co/GAB1vRGl3H
Now out on the cover of @ScienceMagazine:
Our work with @RDasLab on predicting 3D RNA structures!
https://t.co/4iTjfb7G2f
Check out the official press release:
https://t.co/4Faqa0MtTC
and our earlier work on protein complexes:
https://t.co/Aua3f31Ip9
AlphaFold for RNA? Excited to share our @StanfordAILab
work on deep learning for predicting 3D RNA structure, out today on the cover of @ScienceMagazine!
https://t.co/5T9rzHmRP2
Sound interesting? Join us at https://t.co/6Yv5OUqGoX to bring this work to life! More news soon!
We are excited to present our work on Geometric Vector Perceptrons, an equivariant GNN architecture for residue-level protein graphs, at ICLR! Check out our spotlight talk https://t.co/gp8sD0rmIS and Github repo https://t.co/fVriL7ZD9f.
Check out our latest paper exploring the effects of GPCR phosphorylation on arrestin signaling with great collaborators @MattMasureel, Kobilka lab, and @Michel_Bouvier!
We benchmark 3 prototypical architectures - 3D conv. networks, graph networks and equivariant networks - and compare them to 1D/2D baselines. We find that 3D info can strongly improve model performance, but it depends on the choice of architecture for a particular task. [4/4]
Looking for new challenges for machine learning in structural biology?
Check out our recent release: ATOM3D, a unified collection of diverse benchmark datasets for biological problems that deal with atom coordinates in 3D space. โ๏ธโ๏ธโ๏ธ
https://t.co/kmf9lVo1rq [1/4]
The corresponding code to load, filter, and split the ATOM3D datasets is maintained on @github: https://t.co/9Ko3mylkYN.
We hope this lowers the entry barrier for algorithm developers and promotes 3D atomic data as a โmachine learning datatypeโ in its own right. [3/4]
Machine Learning in Structural Biology (@workshopmlsb) is accepted at #NeurIPS2020! Come check out the exciting line-up of speakers and dates for the call for papers at https://t.co/blpB8vVz2M (more details coming soon). Register interest at https://t.co/1Cv0UGB12l
Excited to present our latest work on geometric prediction: the class of prediction problems for (non-scalar) geometric tensors! We show the first real-world demonstration of geometric prediction without the need for scalar approximations. https://t.co/fwUD01NowW [1/n]
Somehow this slipped my radar. Very cool looking work from the @DrorLab: Hierarchical, rotation-equivariant neural networks to predict the structure of protein complexes. https://t.co/Wi97CFMMv8
Excited to share our work on learning from the 3D structure of macromolecules! Our neural network architecture enables us to learn directly from all atoms in protein complexes containing tens of thousands of atoms: https://t.co/LkGNkEIIMH (1/3)
Starting with just the element type of each atom, we learn features at different levels of structural coarseness and aggregate this information hierarchically. The rotation-equivariant network recognizes molecular motifs independent of their orientation. (2/3)
We have 2 papers published in @nature today! ๐ One describes AlphaFold, which uses deep neural networks to predict protein structures with high accuracy. AlphaFold made the most accurate predictions at the 2018 scientific community assessment CASP13. 1/4 https://t.co/TW1zZypsUT