Towards a barrier height benchmark set for biologically relevant systems.

Area of Science:

• Computational chemistry
• Biochemical modeling
• Enzyme catalysis research

Background:

• Accurate computational modeling of enzymatic reactions is crucial for understanding biological processes.
• Existing semi-empirical methods require rigorous benchmarking against high-level theoretical calculations for complex systems.

Purpose of the Study:

• To benchmark the accuracy of semi-empirical methods (PM6, PM7, PM7-TS, DFTB3) against DFT calculations for enzyme reaction barriers and energies.
• To assess the impact of system size, solvation, and geometry re-optimization on the performance of these methods.
• To establish a foundational dataset for a community-driven benchmark of enzymatic reaction modeling.

Main Methods:

• Collected computed barrier heights and reaction energies for five enzyme models from existing literature.
• Utilized B3LYP/6-311+G(2d,2p)[LANL2DZ]//B3LYP/6-31G(d,p) as the reference level of theory.
• Benchmarked PM6, PM7, PM7-TS, and DFTB3, analyzing errors influenced by system size, solvation, and geometry optimization.

Main Results:

• DFTB3 showed a significantly lower mean absolute difference (MAD) of 6 kcal/mol compared to PM6 and PM7 (10-15 kcal/mol) for smaller enzyme models.
• Excluding single outlier systems, PMx and DFTB3 MADs decreased to 4-5 kcal/mol.
• Geometry optimization with PM6 predicted a different reaction mechanism for one system; for others, it increased MADs for large models but decreased them for small models.
• Condensed phase results showed similar accuracy to gas phase results relative to the reference DFT calculations.

Conclusions:

• DFTB3 demonstrates promising accuracy for modeling enzymatic reactions, outperforming PM6 and PM7 in this benchmark study.
• Geometry re-optimization can significantly impact predicted mechanisms and accuracy, highlighting the importance of careful structural considerations.
• The study provides a crucial first step towards a comprehensive, community-contributed benchmark dataset for large enzymatic systems, essential for advancing computational enzymology.