Screening toward the Development of Fingerprints of Atomic Environments Using Bond-Orientational Order Parameters

Area of Science:

• Computational chemistry
• Materials science
• Machine learning

Background:

• Developing efficient molecular representations is crucial for cheminformatics and materials discovery.
• Existing fingerprints may not fully capture both atomic composition and spatial arrangement.
• The QM9 dataset provides extensive molecular properties for computational studies.

Purpose of the Study:

• To identify a concise set of molecular fingerprints that accurately encode atomic and positional information.
• To evaluate the efficacy of these fingerprints in predicting molecular properties.
• To establish a robust descriptor for chemical and structural analysis of organic molecules.

Main Methods:

• Utilized a dataset of 133,885 organic molecules from QM9.
• Generated 504 candidate fingerprints combining atomic numbers and bond-orientational order parameters.
• Employed supervised machine learning regression to screen fingerprints against Open Babel atomic charges.
• Validated fingerprint performance by predicting dipole moments.

Main Results:

• Successfully screened 504 candidate fingerprints down to 60 effective descriptors.
• The selected 60 fingerprints accurately estimate atomic charges derived from Open Babel.
• High accuracy was achieved in predicting dipole moments using these fingerprints.
• Demonstrated the potential of these fingerprints to capture detailed chemical and structural information.

Conclusions:

• A set of 60 fingerprints effectively represents molecular structure and atomic environment.
• These fingerprints offer a promising tool for accurate prediction of molecular properties.
• The findings facilitate advanced computational studies in chemistry and materials science.