3DOpt: Benchmark for Automated Design of 3D Molecular Structures across the Periodic Table.

Area of Science:

• Computational chemistry
• Chemical informatics
• Materials science

Background:

• Automated design of 3D molecular structures is crucial for drug discovery, catalysis, and materials science.
• Existing benchmarks for 3D molecular design methods are limited, primarily focusing on organic compounds.
• A need exists for standardized evaluation across a broader chemical spectrum.

Purpose of the Study:

• To introduce 3DOpt, the first benchmark for assessing 3D molecular generative design across the full chemical spectrum.
• To provide a framework for evaluating methods beyond organic drug-like compounds, including organometallic and transition-metal complexes.
• To establish reference performance metrics for generative molecular design strategies.

Main Methods:

• Development of 3DOpt, a benchmark comprising defined tasks, curated candidate molecule populations, and scoring functions.
• Scoring functions quantify similarity to target 3D structures based on geometry and composition.
• Application of baseline generative methods to demonstrate benchmark utility and establish performance metrics.

Main Results:

• 3DOpt successfully evaluates generative methods across diverse chemical spaces, including organometallic and transition-metal complexes.
• Reference performance metrics were generated for widely used molecular design strategies.
• The benchmark facilitates objective comparison of different 3D molecular generative design approaches.

Conclusions:

• 3DOpt provides a general-purpose framework for the systematic evaluation of 3D molecular generative design.
• This benchmark enables objective assessment of methods across diverse chemical domains.
• 3DOpt will accelerate the development and application of advanced molecular design tools.