AI docking for real-world virtual screening has arrived
London-based startup RECEPTOR.AI announced a new version of its AI docking software, ArtiDock v2.0.
The company also published a preprint disclosing details about the model architecture, training, and tuning.
Receptor.AI notes that the first generation of AI docking techniques used simple and lightweight model architectures and demonstrated results that were inferior to conventional docking.
This issue was mitigated, to some extent, in the second generation of AI docking models by using much heavier architectures. The boost of accuracy, however, came at the expense of very complex architectures, large model sizes, and, as a result, very slow training and inference.
ArtiDock exploits the opposite approach by providing a deliberately lightweight and fast model architecture, which is trained on a larger amount of augmented data. This results in better prediction accuracy without compromising the inference speed.
ArtiDock uses two sources of augmented data: the PocketCFDM algorithm for generating artificial protein-ligand complexes, which mimic real protein binding pockets in terms of statistical distributions of the non-bond interactions, and the ensembles of representative protein conformations obtained from the impressively massive MD simulations of ~17,000 protein-ligand complexes.
ArtiDock 2.0 excels on the PoseBusters v3 dataset, which is deliberately designed to challenge AI docking technologies. The model systematically outperforms all other AI docking techniques, and it also leaves behind conventional docking programs such as Glide, Gold and Vina.
The quality metrics of the generated binding poses are unprecedented for AI techniques and go on par with conventional docking, according to comments from the Receptor.AI team.
All this is several orders of magnitude faster than the industry standard rivals, which allows using ArtiDock in ultra-high-throughput virtual screening scenarios involving multiple protein conformations and multiple explicit off-targets.
In some of the case studies communicated by Receptor.AI, usage of ArtiDock allowed for achieving hit rates of a whopping 40% and discovery of the lead-like compounds with in vivo activity from the single iteration of virtual screening. (link in the comments)
ArtiDock is currently being integrated into the NVIDIA hashtag
#BioNeMo cloud platform for drug development, which will make it available to a wide range of interested biotech and pharma companies in the near future.
#MedicinalChemistry
#DrugDiscovery
#artificialintelligence
#Artidock
Video credit: RECEPTOR.AI