Comparative Exploration on Quantifying Molecular Diversity

Molecular diversity plays a critical role in understanding structure-property relationships, optimizing chemical processes, and improving predictive modeling for extraction, separation, and detection behaviors. Appropriately quantifying molecular similarity and diversity requires effective molecular representations with suitable similarity measures. However, three major research gaps exist: (1) limited understanding of the relationships between structure-based features and physicochemical properties, (2) a lack of studies examining how different binarization approaches and similarity metrics influence similarity assessments, and (3) the absence of a comprehensive evaluation of how different molecular feature spaces impact diversity quantification. This study addresses these gaps by exploring various molecular featurization techniques, including structure-based descriptors (e.g., E-state, Morgan fingerprints, and ABOCH) and physicochemical properties (e.g., Abraham solute descriptors and chromatographic retention factors). We examine how different similarity measures, including distance metrics and binarization techniques, impact the molecular diversity assessment. A novel cumulative binarization approach is introduced to improve the preservation of structural information in similarity calculations. Using a dataset of 109 compounds with experimentally measured retention factors, we compare how different molecular feature spaces influence similarity rankings and diversity quantification. Two additional datasets are also assessed for comparative analysis to further validate the finding. Our findings highlight key differences in structural and physicochemical representations, demonstrating their implications for analytical chemistry and molecular modeling applications.