QM40, Realistic Quantum Mechanical Dataset for Machine Learning in Molecular Science

Area of Science:

• Computational Chemistry
• Materials Science
• Drug Discovery

Background:

• Machine learning (ML) and deep learning (DL) are increasingly used in scientific research.
• A lack of high-quality datasets limits the application of ML/DL for quantum mechanical (QM) predictions.
• Existing datasets do not fully cover the chemical space relevant to drug discovery.

Purpose of the Study:

• To introduce the QM40 dataset, a novel resource designed to address the data scarcity in QM predictions.
• To provide a comprehensive dataset covering a significant portion of FDA-approved drug chemical space.
• To facilitate the development and benchmarking of ML/DL models for QM calculations.

Main Methods:

• Calculated 16 key quantum mechanical parameters for 162,954 molecules using B3LYP/6-31G(2df,p) level of theory in Gaussian16.
• Included molecules with 10 to 40 atoms, focusing on common drug elements (C, O, N, S, F, Cl).
• Ensured compatibility with existing datasets like QM9 and Alchemy for potential concatenation.

Main Results:

• The QM40 dataset represents 88% of the FDA-approved drug chemical space.
• Provides initial and optimized Cartesian coordinates, Mulliken charges, and detailed bond information, including local vibrational mode force constants.
• Offers a robust benchmark for evaluating existing and novel ML/DL methods for QM predictions.

Conclusions:

• The QM40 dataset significantly enhances the availability of high-quality data for ML/DL in computational chemistry and drug discovery.
• It enables more accurate and efficient prediction of molecular properties, accelerating the drug development process.
• Facilitates the advancement of ML/DL techniques for QM calculations by providing a standardized and comprehensive resource.