Using Generative Modeling to Endow with Potency Initially Inert Compounds with Good Bioavailability and Low Toxicity

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Neuroscience

Background:

• Early drug discovery often yields potent compounds with poor drug metabolism and pharmacokinetics (DMPK).
• These developability issues in hit compounds can hinder progression and are difficult to resolve.
• An alternative strategy involves using a
• null library
• of compounds with ideal DMPK properties but no target activity.

Purpose of the Study:

• To explore machine learning for developing drug candidates with desirable DMPK properties from the start.
• To apply generative machine learning to create inhibitors of alpha-synuclein aggregation, a target relevant to Parkinson's disease.

Main Methods:

• Utilized MolDQN, a generative machine learning model.
• Applied the model to impart inhibitory activity against alpha-synuclein aggregation.
• Focused on modifying initially inactive compounds possessing favorable DMPK characteristics.

Main Results:

• Successfully integrated alpha-synuclein aggregation inhibitory activity into compounds with good DMPK profiles.
• Demonstrated the feasibility of using generative modeling for drug developability enhancement.
• Showcased a novel approach to de novo drug design starting from desirable scaffolds.

Conclusions:

• Generative machine learning offers a powerful approach to address DMPK limitations in early drug discovery.
• This strategy enables the creation of drug candidates with both desired potency and developability.
• The method holds promise for accelerating the development of therapeutics for neurodegenerative diseases like Parkinson's.