Receptor.AI Introduces ArtiDock
A new AI docking model designed for high-speed, accurate ligand binding pose prediction
Receptor.AI Introduces ArtiDock
A new AI docking model designed for high-speed, accurate ligand binding pose prediction
PARTNERSHIP
Announcement
Summary
Receptor.AI introduces ArtiDock, an AI docking model that predicts ligand binding poses with high speed and accuracy. By combining data augmentation with a streamlined architecture, ArtiDock outperforms conventional methods like Vina and Gold and demonstrates competitive results compared to newer tools such as AlphaFold. Benchmark results on datasets including Astex and PoseBusters confirm its performance advantages in both predictive quality and inference speed, supporting its application in large-scale virtual screening.
Full Text
Receptor.AI has introduced ArtiDock, an AI docking model designed to predict ligand binding poses with high speed and accuracy. ArtiDock outperforms conventional docking programs such as Vina and Gold and demonstrates competitive performance relative to newer methods, including AlphaFold. For a graphical representation of ArtiDock’s performance on the Astex dataset, refer to Figure 1.
ArtiDock combines data augmentation with a streamlined model architecture, enabling it to recognize a broad range of intermolecular interactions through training on both artificial and real complexes. This approach leads to improved accuracy and predictive robustness. In benchmark tests on the PoseBusters dataset, ArtiDock achieved high predictive precision and was shown to be up to 600 times faster than other models, including recent versions of AlphaFold. Figure 2 illustrates comparative performance on the PoseBusters dataset.
ArtiDock’s operational speed makes it suitable for virtual screening, offering a substantial efficiency advantage. This is visualized in Figure 3, which compares inference times across docking methods. Figure 4 provides a structural quality comparison based on PoseBusters quality metrics.
ArtiDock provides a balance of speed and quality and has already been integrated into Receptor.AI’s virtual screening workflows. Continued development of the model aims to further improve its predictive accuracy and performance across diverse use cases.
