MAPLE: a machine-learning force-field-native platform for automated reaction modeling and enzyme design.

Area of Science:

• Computational chemistry and biology
• Molecular modeling

Background:

• Machine-learning force fields (MLFFs) offer near-quantum mechanical accuracy at reduced computational cost.
• Existing MLFF applications are limited by the lack of unified, automated computational platforms.

Purpose of the Study:

• To introduce MAPLE (MAchine learning Potential for Landscape Exploration), a novel computational toolkit for MLFF-based molecular modeling.
• To provide a unified and automated platform for large-scale and versatile molecular modeling tasks.

Main Methods:

• Developed a tailored software framework and parallelized algorithms for MLFF applications.
• Systematically benchmarked state-of-the-art reactive MLFFs within the MAPLE framework.
• Applied MAPLE to simulate multiple biocatalytic scenarios.

Main Results:

• Demonstrated the robustness and usability of the MAPLE toolkit.
• Highlighted MAPLE's capability for fast and accurate simulation of catalytic reactions.
• Validated MAPLE's performance through systematic benchmarking and application studies.

Conclusions:

• MAPLE integrates accurate MLFFs with parallelized algorithms in an optimized framework.
• MAPLE serves as a next-generation, machine-learning-driven molecular modeling platform.
• MAPLE has broad applicability in rational catalyst design and drug discovery.