DockStream: a docking wrapper to enhance de novo molecular design

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Generative models like REINVENT are crucial for de novo drug design.
• Quantitative structure-activity relationship (QSAR) models enhance target activity prediction but have domain limitations.
• Integrating structure-based methods can overcome QSAR limitations in drug discovery.

Purpose of the Study:

• To introduce a structure-based scoring component for the REINVENT platform.
• To enhance the de novo design of small molecules by overcoming limitations of predictive models.
• To improve the generation of active compounds in drug discovery projects.

Main Methods:

• Implemented DockStream, a molecular docking wrapper, as a scoring component for REINVENT.
• Utilized a benchmarking and analysis workflow within DockStream to automate docking configurations.
• Trained the REINVENT agent using docking configurations informed by public data.

Main Results:

• The integrated structure-based scoring improved REINVENT's ability to retain key binding interactions.
• REINVENT successfully discarded molecules not fitting the binding cavity and utilized sub-pockets.
• Performance was enhanced in scaffold-hopping scenarios, leading to better optimization of docking scores.

Conclusions:

• Structure-based docking via DockStream significantly enhances REINVENT's de novo drug design capabilities.
• This integration overcomes limitations of traditional QSAR models, improving active compound generation.
• The approach optimizes molecular design by considering specific binding site interactions and cavity features.