PM6-ML: The Synergy of Semiempirical Quantum Chemistry and Machine Learning Transformed into a Practical...

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Machine Learning

Background:

• Machine learning (ML) methods are advancing the development of accurate and computationally efficient universal molecular potentials.
• Integrating physical principles or using Δ-ML schemes significantly improves the robustness and transferability of ML models.
• Existing methods often have limitations in chemical space coverage or scalability.

Purpose of the Study:

• To introduce PM6-ML, a Δ-ML method combining the semiempirical quantum-mechanical (SQM) method PM6 with an ML potential.
• To enhance the accuracy, robustness, and transferability of molecular potentials.
• To enable accurate modeling of large biomolecular systems.

Main Methods:

• Developed PM6-ML by synergizing the PM6 SQM method with a state-of-the-art ML potential as a universal correction.
• Applied the Δ-ML scheme within the MOPAC framework.
• Performed extensive benchmarking to assess accuracy and robustness.

Main Results:

• PM6-ML demonstrates superior performance compared to standalone SQM and ML methods.
• The method covers a broader chemical space than previous approaches.
• PM6-ML is scalable to thousands of atoms, enabling applications to large biomolecular systems.

Conclusions:

• PM6-ML offers a significant advancement in accurate and efficient molecular potential development.
• The method's scalability and accuracy make it suitable for complex biomolecular simulations.
• PM6-ML provides a practical tool for computational chemists, with code and parameters readily available.