Integrating multi-algorithm data-driven empirical modeling to unravel structure–antioxidant relationships of medicinal plant polysaccharides

Le résumé fourni par la source

Oxidative stress contributes to the pathogenesis of numerous diseases, and medicinal plant polysaccharides (MPPs) are promising natural antioxidants. However, the substantial structural heterogeneity of MPPs complicates the establishment of reliable structure–activity relationships, thereby limiting their rational screening and development. This study aimed to establish a multi-algorithm, data-driven quantitative structure–activity relationship (QSAR) framework for predicting the antioxidant activities of MPPs. A literature-derived dataset was collected, standardized, and harmonized. Multiple linear regression (MLR), K-means clustering, and principal component analysis (PCA) were used to examine the relationship between polysaccharide molecular weight (MW) and antioxidant activity. MLR showed no clear linear relationship between MW and radical-scavenging activity, suggesting that the association may be nonlinear or driven by multiple structural factors. Clustering analysis further revealed distinct MW-associated distribution patterns, yet antioxidant activity varied widely among polysaccharides with different MW characteristics. Gradient boosting decision tree regression, support vector regression, and random forest (RF) were then used to model the relationship between monosaccharide composition and antioxidant activity. Among the three algorithms, RF performed best, although its generalization capacity remained moderate. Feature-importance and Shapley Additive exPlanations (SHAP) analyses identified several monosaccharides linked to DPPH· and ·OH scavenging activities; these associations were interpreted as predictive rather than causal. External validation with 10 structurally diverse polysaccharides and 8 independent batches of Radix isatidis polysaccharides supported the model-derived relationships. Representative polysaccharides were further evaluated in an H 2 O 2 -induced zebrafish oxidative stress model, and the observed antioxidant effects provided complementary biological support for the model-based screening results. These findings suggest that MW and monosaccharide composition carry useful information for predicting the antioxidant activities of MPPs, although neither descriptor alone fully explains their antioxidant behavior. Combining data-driven modeling, external experimental validation, in vitro assays, and in vivo zebrafish evaluation offers a preliminary framework for exploring structure–antioxidant relationships in MPPs and prioritizing candidates for further study. Larger standardized datasets, more comprehensive structural descriptors, and independent validation are still needed to improve the framework's predictive accuracy, interpretability, and generalizability.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.