Machine learning-driven insights into retention mechanism in IAM chromatography of anticancer sulfonamides: Implications for biological efficacy

Abstract

Machine learning (ML) tools offer new opportunities in drug discovery, especially for enhancing our understanding of molecular interactions with biological systems. This study develops a comprehensive quantitative structure-retention relationship (QSRR) model to elucidate sulfonamides' binding mechanisms to phospholipids via immobilized artificial membrane (IAM) chromatography. Using a dataset of over 500 sulfonamide derivatives, we combined experimental IAM-HPLC data with computational molecular descriptors and ML techniques, achieving robust predictive models. The descriptor-based LASSO regression model effectively predicts retention behavior (R² = 0.71, Q² = 0.77), providing insights into molecular interactions. Critical descriptors influencing these interactions include aqueous solubility, nitrogen-to-oxygen ratio, atomic and mass descriptors such as atom and ring count, as well as logP, indicative of molecular lipophilicity. Furthermore, the fingerprint-based predictive support vector machine model demonstrated superior performance (R² = 0.899 Q² = 0.810) highlighting structural features such as benzene rings and nitrogen-attached fragments as crucial factors in determining phospholipid affinity. Furthermore, predictive models for anticancer activities across three cell lines—HCT-116, HeLa, and MCF-7—were constructed, highlighting CHI_IAM value as a critical determinant of bioactivity. The findings underscore the utility of integrated ML and chromatographic approaches in streamlining the drug development pipeline, improving predictions of biological efficacy while reducing experimental burden.