Autonomous, multiproperty-driven molecular discovery: From predictions to measurements and back

Area of Science:

• Computational Chemistry
• Materials Science
• Organic Synthesis

Background:

• Accelerating the discovery of novel molecules with tailored properties is crucial for advancing various scientific fields.
• Traditional molecular design and synthesis can be time-consuming and resource-intensive.
• Machine learning (ML) offers a promising avenue to enhance the efficiency of molecular discovery.

Purpose of the Study:

• To develop and demonstrate a closed-loop, autonomous molecular discovery platform powered by integrated machine learning tools.
• To accelerate the design and experimental realization of molecules with specific target properties, such as absorption wavelength, lipophilicity, and photooxidative stability.
• To explore the structure-function relationships within diverse and rarely reported molecular scaffolds.

Main Methods:

• Implementation of a closed-loop, autonomous platform integrating machine learning for molecular design.
• Utilizing iterative cycles of molecular design-make-test-analyze (DMTA) for experimental validation.
• Development and application of property prediction models trained on chemical structure-property data.
• Employing multistep syntheses and diverse reaction methodologies for molecule realization.

Main Results:

• The platform experimentally realized 294 previously unreported dye-like molecules across three automated DMTA cycles.
• Exploration of the structure-function space for four rarely reported molecular scaffolds was achieved.
• Property prediction models effectively learned and guided exploration within diverse scaffold derivative spaces.
• A second study successfully identified nine high-performing molecules in a less-explored chemical space using trained models.

Conclusions:

• The developed autonomous platform significantly accelerates the discovery of molecules with desired properties.
• Integrated machine learning tools are effective in navigating complex structure-property landscapes.
• The platform demonstrates versatility in exploring both well-studied and underexplored chemical spaces for novel molecule generation.