Correlation-Based Framework for Extraction of Insights from Quantum Chemistry Databases: Applications for...

Area of Science:

• Computational Chemistry
• Materials Science
• Data Mining

Background:

• Growing volume of quantum chemistry (QC) data necessitates advanced analysis methods.
• Current atom-level understanding often relies on manual data analysis.
• Need for automated tools to extract insights from complex molecular datasets.

Purpose of the Study:

• To develop and present a data mining framework for accelerated insight extraction from QC datasets.
• To investigate the relationship between atomic features and molecular properties using correlation analysis.
• To provide a practical Python package for QC data mining.

Main Methods:

• Featurization of atomic data into molecular properties (AtoMF).
• Application of correlation coefficients (Pearson, Spearman, Kendall) to analyze feature-property relationships.
• Testing the framework on three distinct nanocluster systems: PtTM_55-, CeZr_15-O₃₀, and (CH + mH)/TM₁₃.

Main Results:

• Spearman and Kendall correlation coefficients proved effective for identifying consistent insights in nanocluster systems.
• Pearson correlation coefficient demonstrated high sensitivity to outliers, rendering it unsuitable for this analysis.
• Identified specific atomic features correlating with target properties in the studied nanoclusters.

Conclusions:

• The proposed data mining framework effectively accelerates the extraction of knowledge from QC data.
• Spearman and Kendall correlations are recommended for analyzing QC data due to their robustness.
• The developed Python package offers a valuable tool for researchers in computational chemistry and materials science.