Lasso regression with molecular descriptors predicts natural product adsorption onto graphene oxide-derived adsorbents

Natural products and their derivatives are critical resources for drug discovery, with their pharmaceutical applications requiring efficient extraction and purification methodologies. In recent years, the emergence of sustainable development paradigms and advancements in naturally derived apt nanomaterials bring in graphene composites, featuring unique two-dimensional architectures and tunable surface chemical properties, as ideal adsorbents to amplify natural product processing efficiency. In this study, three graphene derivatives—graphene oxide (GO), carboxylated graphene oxide (GO-COOH), and tetraethylenepentamine-functionalized reduced graphene oxide (rGO-TEPA), were employed as adsorbents for the green extraction and purification of 28 types of structurally diverse natural products. By developing a LASSO regression model incorporating 1332 molecular descriptors, quantitative structure-adsorption relationships (QSARs) were established. Their critical descriptors (R7m, nRCOOH, and E3s) were identified to elucidate the significant influence of atomic polarizability, electronegativity, and hydrogen bonding on adsorption mechanisms. Structure-adsorption correlation analyses demonstrate that this machine learning framework facilitates high-throughput screening of natural product libraries using graphene-based adsorbents. This methodological advancement promises to enhance natural product utilization and streamline small-molecule drug discovery through rational adsorbent selection.