Freedom Space 3.0: ML-Assisted Selection of Synthetically Accessible Small Molecules.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Machine learning (ML) significantly advances chemical space exploration.
• ML enables the creation of tailored molecular subsets for specific applications.
• Developing novel, synthetically feasible chemical libraries is crucial for drug discovery.

Purpose of the Study:

• To develop Chemspace Freedom Space 3.0, a large chemical library of synthetically accessible small molecules.
• To utilize ML-based filtering of building blocks for enhanced molecule quality and synthetic feasibility.
• To provide a complementary resource to existing chemical spaces like Enamine REAL Space.

Main Methods:

• Developed a custom molecular representation for ML model training.
• Applied ML filtering to refine building block selection prior to molecular enumeration.
• Generated a library of 5 billion molecules using ten validated chemical transformations.
• Computationally evaluated physicochemical properties, chemical diversity, and synthetic accessibility.

Main Results:

• Freedom Space 3.0 contains 5 billion synthetically feasible small molecules.
• Computational analysis confirmed favorable physicochemical properties and chemical diversity.
• Experimental validation achieved over 80% synthesis success rate within 4-6 weeks for 700 molecules.

Conclusions:

• Chemspace Freedom Space 3.0 is a high-quality, synthetically accessible chemical library.
• The ML-driven approach enhances the quality and feasibility of generated molecules.
• Freedom Space 3.0 has the potential to accelerate hit finding and follow-up in drug discovery workflows.