[关键词]
[摘要]
目的 在脑缺血再灌注损伤(cerebral ischemia-reperfusion injury,CIRI)大鼠模型中,基于可解释性机器学习探讨丹参注射液的多成分药动学-药效学(pharmacokinetic-pharmacodynamic,PK-PD)关系以及各成分的药效贡献度。方法 建立CIRI大鼠模型,于再灌注后立即尾iv丹参注射液(2.10 mL/kg)。采用高效液相色谱法(high-performance liquid chromatography,HPLC)测定血浆中原儿茶酸(protocatechuic acid,PAD)、原儿茶醛(protocatechuic aldehyde,PAE)、丹酚酸A(salvianolic acid A,SAA)和丹酚酸B(salvianolic acid B,SAB)的质量浓度,采用ELISA检测血浆中超氧化物歧化酶(superoxide dismutase,SOD)水平,以0 h个体基线校正后的SOD变化量(△SOD)作为外周抗氧化PD指标。采用神经常微分方程(neural ordinary differential equation,Neural ODE)表征多成分PK关系,构建遗传神经网络(genetic algorithm with back propagation neural network,GA-BPNN)描述PK-PD非线性关系,并结合沙普利加性解释(Shapley additive explanations,SHAP)方法量化各成分对药效的贡献度。结果 HPLC结果显示,PAD、PAE、SAA与SAB在给药后均呈现清晰的质量浓度-时间动态变化特征。且SOD在丹参注射液干预后呈早期升高、随后逐渐趋于稳定的动态变化。针对PK建立模型,Neural ODE对4种成分的拟合度分别为0.965、0.994、0.986和0.966;针对PK-PD建立模型,GA-BPNN模型拟合度达到0.971。SHAP分析表明,各活性成分对SOD响应的贡献存在显著差异,整体贡献度排序为PAE>SAB>PAD>SAA。同时根据SHAP结果,提示丹参注射液的药效响应可能由多成分协同驱动。结论 丹参注射液在大鼠中表现出显著增强的抗CIRI作用,Neural ODE可较好表征多成分PK关系,GA-BPNN可较好表征非线性PK-PD关系。
[Key word]
[Abstract]
Objective To investigate the multi-component pharmacokinetic-pharmacodynamic (PK-PD) relationships and pharmacodynamic contribution profiles of Danshen Injection (丹参注射液, DI) in a rat model of cerebral ischemia-reperfusion injury (CIRI) based on explainable machine learning approaches. Methods A CIRI rat model was established, and DI (2.10 mL/kg) was administered intravenously via the tail vein immediately after reperfusion. High-performance liquid chromatography (HPLC) was used to determine the plasma concentrations of protocatechuic acid (PAD), protocatechuic aldehyde (PAE), salvianolic acid A (SAA), and salvianolic acid B (SAB). The level of superoxide dismutase (SOD) was detected by ELISA, and the change in SOD (△SOD) after baseline correction of individual values at 0 h was used as the peripheral antioxidant PD index. Neural ordinary differential equations (Neural ODE) were applied to characterize the multi-component PK profiles, while a genetic algorithm with back propagation neural network (GA-BPNN) was constructed to describe the nonlinear PK-PD relationships. Shapley additive explanations (SHAP) analysis was further employed to quantify the contribution of each component to the pharmacodynamic response. Results HPLC results demonstrated that PAD, PAE, SAA and SAB exhibited distinct concentration-time dynamic profiles after administration. Moreover, SOD exhibited a dynamic pattern of early elevation followed by a gradual stabilization after DI intervention. For PK modeling, the fitting coefficients of the Neural ODE model for the four components were 0.965, 0.994, 0.986 and 0.966, respectively. For PK-PD modeling, the GA-BPNN model achieved a fitting coefficient of 0.971. SHAP analysis revealed significant differences in the contributions of the active components to the SOD response, with the overall contribution ranking of PAE > SAB > PAD > SAA. Moreover, the SHAP results suggested that the pharmacodynamic response of DI may be jointly driven by multi-component synergistic effects. Conclusion DI exhibited a significantly enhanced anti-CIRI effect in rats. Neural ODE was able to adequately characterize the multi-component PK relationships, and GA-BPNN could effectively capture the nonlinear PK-PD relationships.
[中图分类号]
R285.61
[基金项目]
浙江省自然科学基金项目-优青项目(LZYQ25H270001);浙江省自然科学基金项目-探索一般(LY24H270007); 浙江省中医药管理局中医药健康服务研究计划项目(2026ZF41)