基于随机森林回归模型的中药复方金复康对肺癌细胞增殖作用的组方优化

Optimization of Effect of Traditional Chinese Medicine Compound Jinfukang on Lung Cancer Cell Proliferation Based on Random Forest Regression Model

目的:以治疗肺癌疗效确切的中药复方金复康为研究对象,从抑制肺癌细胞增殖方面对其处方组成进行优化,并为中药复方的处方优化提供新的途径。方法:应用筛选试验设计(Plackett-Burman),然后应用随机森林算法,建立基于设计处方集的预测模型,61个组方设计矩阵为输入,以相应的61个组方的半数抑制浓度(half maximal inhibitory concentration,IC50)的自然对数为输出建立随机森林回归模型。并用5次10折交互验证优化模型参数mtry,最后应用网格化搜索算法得到最优的配伍组,并进行肺癌细胞增殖实验验证,以金复康组成药物的不同组合作用于肺腺癌细胞株A549的IC50为指标,从抑制肺癌细胞增殖方面对金复康的组方配伍进行优化。结果:通过随机森林模型结合网格化搜索算法得到的最优配伍组为黄芪、麦冬、重楼、女贞子和绞股蓝,且经实验验证,在抑制肺癌细胞增殖方面,金复康优化方相对于金复康原方具有更好的抑制细胞增殖效应。结论:基于随机森林模型结合网格化搜索算法能够为复杂性中药复方的组方优化提供方法学参考。

Objective： To study the accurate curative effect of traditional Chinese medicines（ TCM）compound Jinfukang in treating lung cancer,and optimize the prescription composition to inhibit lung cancer cell proliferation,so as to provide a new approach for the optimization of TCM compound prescription. Method：Screening test design（ Plackett-Burman） and random forest algorithm were applied in establishing the prediction model based on design prescription set. With 61 composition design matrixes as the input,and natural logarithm of their corresponding IC50 as the output,random forest regression model was establish. The optimization model parameter mtry was verified by 5-time and 10-fold interaction. Finally,a grid search algorithm was applied to get the optimal group,and verify the lung cancer cell proliferation experiment. With the IC50 obtained from different formula of Jinfukang on A549 lung adenocarcinoma cancer cells as an index,the composition of Jinfukang were optimize to inhibit lung cancer cell proliferation. Result： Through random forest model combined with grid search algorithm,we got the optimal compatible herbs,namely Astragali Radix,Ophiopogonis Radix,Paridis Rhizoma,Ligustri Lucidi Fructus and Gynostemmatis Pentaphylli Herba seu Radix. Through the experiment,optimized Jinfukang showed a better effect in cell inhibiting and proliferation than original Jinfukang in the aspects of inhibition of lung cancer cell proliferation. Conclusion： The random forest model combined with the grid search algorithm complexity can provide methodology reference for the optimization of composition of complex TCM compound formula.