We're entering a phase where AI and physics no longer compete in molecular AI models; they collaborate. DiffDock-Glide: a hybrid model that guides AI to real binding sites, and then lets physics take over. It beats Glide on AF2 structures and even docks to allosteric sites.

Try DiffDock on 310 Copilot here: 3 min tutorial here:

Congratulations to Gabriele Corso for successfully defending his PhD thesis! From DiffDock to Boltz-1, it's been an incredible journey of exciting breakthroughs in molecular machine learning and we can't wait to see what the future holds for Gabriele 🔥 ⚗️ 💻

🥇 We find that current performance estimates are not only inflated by up to 2-3x, but surprisingly & excitingly, also that deliberate splits + clustering allows training models that generalize much better! Via extensive benchmarking and retraining of DiffDock(-PP), enabled by

Interesting blind docking results from Vina / SMINA / GNINA seem to perform well here compared to DiffDock.

ok i think the diffdock people won

Very cool to see Jensen Huang, once again, highlight DiffDock as a key component of the @nvidia #BioNeMo platform for virtual screening in his much-anticipated keynote at #GTC24!

NVIDIA's newly announced open-source GPU-accelerated, BioNeMo Framework speeds up drug discovery and molecule design by combining AI model scaling, optimized inference, and automated workflows. → AlphaFold2 on NIM achieves 5x faster protein structure predictions. DiffDock 2.0

v1.1 DiffDock

DiffDock-Glide: a hybrid physics-based and data-driven approach to molecular docking. 1． DiffDock-Glide integrates deep learning with traditional physics-based methods for improved molecular docking accuracy. 2． It combines DiffDock's generative model for pose sampling with

Love to see users create videos of the BioNeMo platform! Here’s a YouTube video posted by a user testing out ESMFold, MolMIM and DiffDock using publicly available data!

all the bio kids want to pull up to the bioML party until they read the diffdock paper and realize they need to learn non-convex manifold optimization to understand methods from 3 years ago

New updates on DiffDock + accepted to #ICLR!.Knowing how small molecules bind to proteins is important for drug discovery: in the crucial scenario where the bound protein structure is unknown DiffDock also outperforms the baselines (10.4% success vs 21.7%)

interesting to see Schrödinger take an anti-ML view in their investor materials. this is an antiquated view of ML (modern ML is capable of extrapolation vs interpolation). recent demonstrations such as DiffDock (diffusion model for molecular docking) show that ML is advancing

Very excited to share DiffDock-PP, a new diffusion generative model that builds upon the recently proposed DiffDock and adapts it for the task of rigid-body protein-protein docking. This work was accepted to the ICLR MLDD 2023 workshop. Paper: 1/4

Excited to introduce FeatureDock, a docking software that uses a Transformer encoder trained on local chemical features to predict ligand binding. FeatureDock outperforms DiffDock, Smina, and AutoDock in distinguishing strong inhibitors from weak ones.

DEEP-LEARNING BASED DOCKING METHODS: FAIR.COMPARISONS TO CONVENTIONAL DOCKING WORKFLOWS. 1. DiffDock, a deep-learning-based molecular docking method, shows nominally better results compared to traditional methods in its reported tests. However, this study challenges those claims

SN25 is rolling out Protein-Ligand Docking in our soon-to-be-released product frontend using DiffDock🤝 . Take stable molecular dynamics solutions from the miners, and effortlessly run multi-conformation inference🧬. @MacrocosmosAI

DiffDock-Glide: a hybrid physics-based and data-driven approach to molecular docking 結合部位にバイアスをかけたDiffDockでポーズを出してGlideでポーズ/スコアを精緻化.1stじゃないがシュレの人も共著.

Diffdock & ESMFold are now 6x & 7x faster!.With 310 Copilot + Nvidia NIMs. Try it for free here .