Data and Molecular Fingerprint-Driven Machine Learning Approaches to Halogen Bonding.

Area of Science:

• Computational Chemistry
• Machine Learning in Chemistry
• Supramolecular Chemistry

Background:

• Predicting halogen bond (XB) strength and donor properties is crucial for medicinal chemistry and materials science.
• Current methods like ab initio calculations are computationally expensive.
• There is a growing need for fast, accurate, and efficient prediction tools.

Purpose of the Study:

• To develop and apply machine learning models for predicting halogen bond properties.
• To classify XB donors and complexes based on their principal halogen atom.
• To predict the electrostatic potential (Vs,max) and interaction strength of XB complexes.

Main Methods:

• Utilized three machine learning models.
• Employed molecular fingerprint and data-based analysis.
• Compared predictions with density functional theory (DFT) calculations.

Main Results:

• Fingerprint analysis yielded root-mean-square errors of ~7.5 and ~5.5 kcal mol⁻¹ for Vs,max prediction in halobenzene and haloethynylbenzene systems.
• Binding energy predictions for XB donors and ammonia acceptors were within 1 kcal mol⁻¹ of DFT-calculated energies.
• Precalculated DFT data led to more accurate predictions than fingerprint analysis.

Conclusions:

• Machine learning models provide a viable and efficient method for predicting halogen bond properties.
• The developed models show promise for accelerating research in medicinal chemistry and materials science.
• Data-driven approaches offer a valuable complement to traditional computational chemistry methods.