基于人工智能GPGA的华法林剂量预测模型

The warfarin dose predictive model based on artificial intelligence GPGA

目的提出一种基于遗传程序设计(GP)和遗传算法(GA)的人工智能算法,命名为遗传程序设计-遗传算法(GPGA),以建立更准确的华法林剂量预测模型。方法纳入100例华法林达稳态,临床信息完整,且获得CYP2C9与VKORCI基因型的汉族患者,记录其华法林稳定维持剂量(真实值)。使用GPGA算法合成并进化模型得出预测的华法林维持剂量(预测值),与本院提出的线性回归模型、目前国际公认的国际华法林药物基因组学联合会IWPC模型,及三种现有人工智能建模方法相比较。结果在各复杂度模型中,GPGA的平方相关系数(R2)、均方误差(MSE)和预测值在真实值±20%范围内的比例(20%-p)总体表现最优。GPGA得到的R2从训练集到测试集没有下降。身高和性别变量的加入并未进一步提高模型预测性能。结论 GPGA总体得到了最好的趋势相关性、精度和可用性,且泛化性强。身高和性别对华法林维持剂量无明显预测价值。

Objective To develop a new artificial intelligence method,based on Genetic Programming（GP）and Genetic Algorithm（GA）,denoted by Genetic Programming-Genetic Algorithm（GPGA）,and to improve the accuracy of warfarin dose predictive model.Method Clinical data,CYP2 C9 and VKORC1 genotypes and warfarin maintenance dose（actual data）of 100 Chinese Han patients undergoing warfarin therapy with stable international normalized ratio（INR）were collected in the First Affiliated Hospital of Soochow University from January 2014 to February 2017.Utilizing GPGA,we generated and evolved warfarin dose prediction models,and got the predicted warfarin dose（predicted data）.Then we compared our models with a linear regression model developed by our hospital earlier,the internationally acknowledged model developed by the International Warfarin Pharmacogenetics Consortium（IWPC）,as well as three existing artificial intelligence algorithm.Results Among all the models with different complexities,GPGA gave the best overall performance in the squared correlation（R2）,mean square error（MSE）and the percentage of patients whose predicted dose of warfarin were within ±20% of the actual dose（20%-p）.Besides,GPGA had no reduction on R2 from training set to test set.Height and gender did not improve the performance of the model.Conclusions GPGA developed in this study shows an overall best correlation,accuracy and applicability among all the models mentioned above.Besides,GPGA shows a good extensiveness for new data.Height or gender is not predictive for the maintain dosage of warfarin.