QuanDB: a quantum chemical property database towards enhancing 3D molecular representation learning

Area of Science:

• Computational Chemistry
• Materials Science
• Drug Discovery

Background:

• Molecular 3D structure and electronic properties are vital for predicting chemical behavior and interactions.
• Existing quantum chemical (QC) property datasets may lack comprehensive structural diversity or user accessibility.
• Accurate QC data is essential for advancing computational chemistry and machine learning applications in molecular design.

Purpose of the Study:

• To develop a high-quality, comprehensive quantum chemical (QC) property database named QuanDB.
• To provide stable 3D molecular conformations and a wide range of QC properties for diverse chemical entities.
• To create a user-friendly resource for machine learning-based molecular design and chemical space exploration.

Main Methods:

• Compilation of 154,610 compounds from public databases and literature, covering 10,125 scaffolds and nine elements.
• Calculation of 53 global and 5 local QC properties, including geometric structure, electronic structure, and thermodynamics.
• Utilized B3LYP-D3(BJ)/6-311G(d)/SMD/water for geometry optimization and B3LYP-D3(BJ)/def2-TZVP/SMD/water for single point energy calculations.

Main Results:

• Developed QuanDB, a database featuring 154,610 molecules with detailed QC properties and stable 3D conformations.
• The database includes 10,125 unique scaffolds and covers elements H, C, O, N, P, S, F, Cl, and Br.
• Calculations required over 107 core-hours, ensuring high accuracy for the provided geometric and electronic structure data.

Conclusions:

• QuanDB offers valuable data for molecular representation models crucial for machine learning-driven molecular design.
• The database serves as a benchmark for training and optimizing machine learning models in chemistry.
• QuanDB is freely accessible and poised to accelerate the discovery of novel drugs and materials.