人工神经网络模型在肾移植患者西罗莫司个体化给药中的应用

目的:建立人工神经网络用于估算西罗莫司血药浓度的方法。方法:收集56例肾移植患者口服西罗莫司的182份全血浓度数据,采用遗传算法配合动量法优化网络参数,建立人工神经网络,并对测试数据进行处理,验证测试结果。结果:人工神经网络平均预测误差(MPE)与平均绝对误差(MAE)分别为(0.31±1.14)、(0.89±0.77)ng·mL-1,32例/次(88.9%)血药浓度数据绝对预测误差≤2.0ng·mL-1。人工神经网络模型准确性及精密度优于多元线性回归及非线性混合效应模型。结论:人工神经网络模型可用于预测西罗莫司血药浓度,指导个体化给药。

Application of Artificial Neural Network in Individual Medication of Sirolimus for Renal Transplantation Recipients

OBJECTIVE:To establish a prediction method for blood concentration of sirolimus by artificial neural network.METHODS:182 blood concentration samples were collected from 56 renal transplantation recipients.Artificial neural network(ANN) was established after network parameters were optimized by using momentum method combined with genetic algorithm.The trial data was processed and validated.RESULTS:Mean prediction error(MPE) and mean absolute error(MAE) of ANN were(0.31±1.14)ng·mL-1 and(0.89±0.77)ng·mL-1,respectively.The absolute prediction error of testing data of 32 cases/times(88.9%) was less than 2.0 ng·mL-1.Accuracy and precision of ANN were superior to multiple regression and nonlinear mixed effect model.CONCLUSION:ANN is suitable to predict blood concentration of sirolimus and guide individual medication.