Chemistry-Informed Machine Learning Framework for Predicting Structural Properties in Osmabenzene Complexes.

Area of Science:

• Organometallic Chemistry
• Computational Chemistry
• Materials Science

Background:

• Metallabenzenes, metal-containing aromatic compounds, show significant out-of-plane distortion.
• This non-planarity deviates from classical aromaticity rules but enhances aromatic character.
• The phenomenon is explained by a sigma-orbital antibonding interaction, termed the sigma-control mechanism.

Purpose of the Study:

• Investigate the structural properties of osmabenzene complexes.
• Apply a chemistry-informed machine learning approach to predict non-planarity.
• Validate the sigma-control mechanism and identify key factors influencing distortion.

Main Methods:

• Compiled a dataset of 329 osmabenzene structures.
• Developed mechanistically guided descriptors from coordination chemistry and molecular orbital theory.
• Employed machine learning models, including those with orbital energy descriptors and ligand-level descriptors, alongside SHAP analysis.

Main Results:

• Machine learning models with orbital energy descriptors achieved high predictive accuracy, confirming the sigma-control mechanism.
• A chemically interpretable set of ligand-level descriptors maintained high performance (R² = 0.970, RMSE = 1.990°, MAE = 1.544°).
• SHAP analysis identified axial ionization potential and equatorial polarizability as primary drivers of non-planarity.

Conclusions:

• Demonstrated the efficacy of integrating mechanistic chemical insights into data-driven modeling.
• Provided a practical framework for predicting and designing functional metalla-aromatic materials.
• Highlighted the importance of the sigma-control mechanism in metallabenzene structural properties.