“区块链”融合深度学习的中药上市后肾损伤预警模型研究

中药安全性始终被重视，中药注射剂的过敏反应和口服中药对重要脏器的安全性一直是重点。项目团队已完成中药致过敏反应、肝损伤等风险信号研究，拟进行中药致肾损伤风险信号和效应机制预测。基于已构建的400万例HIS数据仓库、SRS数据库等，利用“区块链”去中介化特点的思想，通过哈希算法生成特定摘要值，实现多源异构数据的整合与分布式存储；采用Group Lasso筛选变量，采用卷积神经网络、深度置信网络逐层训练算法等深度学习的方法构建预警模型，解决空间参数少、危险因素亟待深层优化的需求，采用ROC曲线判别预警模型等方法，进行肾损伤风险信号挖掘；同时，进行中药致肾损伤靶点通路复杂生物网络预测，实现含多隐层的中药对肾脏安全性宏观预警信号和微观效应机制的激活数据学分析，完善中药对重要脏器安全性多源证据，也为中医药科研引入“块数据“思想探索新方法。

The safety of traditional Chinese medicine has always been emphasized. The allergic reaction of traditional Chinese medicine injections and the safety of oral Chinese medicine to important organs have always been the focus. The project team has completed the study of risk signals such as allergic reaction and liver injury caused by traditional Chinese medicine, and plans to predict the risk signals and effect mechanism of kidney injury caused by traditional Chinese medicine. Based on the 4 million HIS data warehouses and SRS databases that have been constructed, using the idea of "block chain" to de-mediate, the specific summary values are generated by hashing algorithm to realize the integration and distributed storage of multi-source heterogeneous data. Group Lasso is used to filter variables, and convolutional neural network and deep confidence network are used to train layer by layer. The method of deep learning such as algorithm is used to construct early warning model to solve the need of less spatial parameters and urgent optimization of risk factors. The ROC curve discriminant early warning model is used to mine the risk signal of kidney injury. At the same time, the complex biological network of target pathway of kidney injury caused by traditional Chinese medicine is predicted to realize the multi-hidden layer of traditional Chinese medicine. The activation data analysis of macro-warning signal and micro-effect mechanism of kidney safety can improve the multi-source evidence of the safety of traditional Chinese medicine to important organs, and also introduce the idea of "block data" into the research of traditional Chinese medicine to explore new methods.