ChemSpaceAL: An Efficient Active Learning Methodology Applied to Protein-Specific Molecular Generation.

Area of Science:

• Computational chemistry
• Artificial intelligence in medicine
• Drug discovery and development

Background:

• Generative artificial intelligence (AI) models offer powerful capabilities for applications in drug discovery.
• Exploring vast chemical spaces requires efficient methods to identify molecules with desired properties.
• Active learning strategies can optimize the search for novel molecular structures.

Purpose of the Study:

• To present a computationally efficient active learning methodology for targeted molecular generation.
• To demonstrate the applicability of this methodology in identifying potential drug candidates.
• To provide an open-source software package for reproducibility and implementation.

Main Methods:

• Development of a novel active learning framework for molecular generation.
• Application of the methodology to c-Abl kinase, a target with known inhibitors.
• Testing the method on the HNH domain of the CRISPR-associated protein 9 (Cas9) enzyme, a target lacking known inhibitors.

Main Results:

• The active learning model successfully generated molecules similar to known c-Abl kinase inhibitors without prior knowledge.
• The model reproduced two existing FDA-approved small-molecule inhibitors of c-Abl kinase exactly.
• The methodology proved effective for generating molecules targeting the Cas9 enzyme's HNH domain.

Conclusions:

• The presented active learning methodology is computationally efficient and effective for targeted molecular generation in drug discovery.
• This approach can identify novel drug candidates for proteins with and without existing inhibitors.
• The open-source ChemSpaceAL package facilitates the implementation and reproducibility of this AI-driven drug discovery technique.