Application of modified partial least squares in data analysis of Traditional Chinese Medicine

摘要

Because of the complexities of traditional Chinese medicine's prescriptions, the dose-effect relationship between prescriptions has a significant difference from the common “S” - type curve of the pharmaceutical chemicals, which is nonlinear. Therefore, the study of the dose-effect relationship between prescriptions can not copy the research methods of dose-effect relationship of pharmaceutical chemicals, but need to consider a variety of influencing factors and compatibility of medicines. Based on the collection, collation and analysis of experimental data in a large number of literature of Traditional Chinese Medicine (TCM) prescriptions, this paper first planned to construct algorithm which fused Q-type clustering and R-type clustering to eliminate abnormal data; obtain high-efficiency modeling samples through the correction method of orthogonal signal; build a complete path graph by making the respective variables and dependent variables as nodes and using direct and indirect path coefficient as weights, and analyze the directional and authoritative graph through the complex network model to filter the important variables out; The partial least squares (PLS) nonlinear model towards the dose-effect relationship of TCM was established based on the maximum entropy principle to determine the partial least squares, which has great significance to scientifically illustrate the dose-effect relationship between prescriptions and its effects, systematically study, summarize and draw theories of prescriptions' doses, rationally improve the clinical effects of TCM and guide the choices of clinical doses.