Recognizing molecular patterns by machine learning: an agnostic structural definition of the hydrogen bond

Area of Science:

• Computational Chemistry
• Materials Science
• Chemical Physics

Background:

• Chemical bonding rationalizes recurring structural patterns in molecules and solids.
• Hydrogen bonds are crucial in water, biological systems, and materials, but defining them is challenging due to their variable nature.
• Existing methods for hydrogen bond identification rely on complex electronic structure calculations or arbitrary structural parameter choices.

Purpose of the Study:

• To develop an algorithmic definition of a chemical bond based solely on structural information.
• To introduce a univocal, unbiased, and adaptive method for identifying hydrogen bonds.
• To demonstrate the application of machine learning for pattern recognition in atomistic simulations.

Main Methods:

• Utilizing machine learning analysis of atomistic simulations.
• Developing an automated system for identifying atomic patterns.
• Establishing an algorithmic definition of a bond based on structural data.

Main Results:

• Successfully identified atomic patterns automatically.
• Provided an unbiased and adaptive definition for hydrogen bonds.
• Demonstrated a strategy adaptable to recognizing structural patterns in diverse chemical compounds.

Conclusions:

• The proposed machine learning strategy offers a robust method for defining chemical bonds algorithmically.
• This approach provides an unbiased and adaptive definition for hydrogen bonds, overcoming limitations of previous methods.
• The methodology can be extended to identify and classify various structural patterns in materials and chemical systems.