融合深度学习的偏最小二乘法在中医药数据分析的研究

“TCM does not pass the secret lies in the dose”, Chinese medicine is a treasure of traditional Chinese medicine, Analysis relationship between the amount of medicine and effect is the safe, effective premise and foundation of clinical application. However, dose-effect relationship between the components of the data is not clear medicine complex than chemical drugs. Need to adapt to the characteristics of traditional Chinese medicine、can handle high-latitude、 low DOF、missing values and more quantitative indicators and qualitative indicators mixed patterns and techniques of data mining system.The projects undertaken by the State to organize and extract of Chinese medicine prescription medicine major issue on the basis of experimental data; Improved OSC method for removing the independent variables and the dependent variable in the orthogonal noise information, access to high-quality modeling sample; The independent variables and the dependent variable node、the direct and indirect path coefficients as weights to build a complete path diagram, A typical use of complex network analysis algorithms have analyzed this chart to the right to filter out the important variables; Improved integration CNN algorithms to extract effective principal component, Embed Softmax regression, The eventual establishment of partial least squares nonlinear regression models for large data medicine dose relationship. The project aims to explain the science of Chinese medicine dose-response relationship and its influencing factors、provide the system of traditional Chinese medicine dose theory the technical support. To provide new ideas of the accuracy of data mining、stability and generalization.

“中医不传之秘在于药量”，中医药学是我国传统医学的瑰宝，分析中药用量与效果间的关系是临床应用安全、有效的前提和基础。但效应成分不清晰的中药的量效数据关系比化学药复杂，亟需适应中药特点的，可以处理高纬度、低自由度、缺失值多、定量指标和定性指标混杂的数据挖掘模式和技术方法体系。本项目在整理和抽取承担的国家中医药重大课题中医方药实验数据的基础上；改进OSC方法去除自变量中与因变量正交噪音信息，获取高质量的建模样本；将自变量与因变量为节点、直接和间接通径系数为权值构建完整的通径图，利用复杂网络典型分析算法对此有向有权图进行分析筛选出重要变量；融合改进CNN算法来提取有效主成分，嵌入Softmax回归，最终建立面向中药量效大数据关系的偏最小二乘非线性回归模型。本项目旨在为科学阐释中药量效关系及其影响因素、系统研究中药剂量理论提供技术支撑。为提高中医药数据挖掘的精度性、稳定性、泛化性探索新的思路。

科学挖掘分析中药的用量与效果间的关系是临床应用安全、有效的前提和基础，中医药数据分析模式尚不成熟，亟需适合其特点的技术方法体系。本课题首先抽取、整理实验与文献的中药量效实验数据；采用灰色关联法方法，剔除自变量中与因变量噪音信息，获取高质量的建模样本；将各自变量与因变量为节点、直接和间接通径系数为权值构建完整的通径图，根据路径综合权重值筛选出对因变量直接和间接影响大的自变量点群；同时利用深度学习模型抽取原始数据的上层特征，将抽取的特征放入偏最小二乘模型进行多元线性回归，精准选取主成分，最终形成面向中药量效大数据关系的偏最小二乘优化模型。研究结果将对科学地阐释中药量效关系及其影响因素，系统地研究、总结和提炼中药剂量理论，指导临床合理选择剂量，提高中医的临床疗效具有一定意义。在研究过程中，共发表论文16篇，分别有SCI期刊7篇，EI论文3篇，中文核心期刊5篇，普通期刊1篇，以及发明专利1项，实用新型专利3项。

内容获取失败，请点击重试