Study On The Multi-objective Optimization Method For Dose-effect Relationship Of Traditional Chinese Medicine

In this dissertation, we mainly studied the multi-objective optimization methodsfor the dose-effect relationship of traditional Chinese medicine in proceeding theprevious experimental data of the active fraction of Rhubarb on treatment of ischemiccerebral stroke. First, the dose-effect relationship was established by three kinds ofmodels, including random forest regression (RFR), support vector machine (SVM),and radial basis function artificial neural network (RBF-ANN). After a comparison itwas found that the RBF-ANN performed the best and was thus selected to conduct thenext multi-objective optimization research. Secondly, the multi-objective optimizationmodels based on weight modification method, fractional programming method andNSGA-II were built to find the optimal combination of the Rhubarb fractions, whoseparameters were optimized by the adaptive genetic algorithm. The main researchcontents were as follows:1. Construction and selection of the mathematical modelsIn this chapter three methods were used to build the dose-effect relationshipmodels. First, the different combinations of the five compounds from Rhubarb wereused as the input variables and the six pharmacological indices were used as theoutput variables. The parameters of each model were selected by the rule of thesynthesis function of fractional programming method and the mean square error andcorrelation coefficients were calculated by the adaptive genetic algorithm. TheLeave-one-out Validation method was used to assess the model accuracy. From theresults it was demonstrated that the predictive effect of RBF-ANN was the best, withthe mean square errors being [25.91670.53540.05530.00250.027033.5496], and thecorrelation coefficients being [0.74630.89990.87060.99830.95540.7543]. Therefore, the RBF-ANN models were chosen to conduct the next multi-objectiveoptimization research.2. Traditional methods for multi-objective optimizationIn this chapter，two common traditional multi-objective optimization methods fordose-effect relationship of traditional Chinese medicine, namely weight modificationmethod and fractional programming method, were introduced. The weightingmodification method was divided into subjective weighting method, objectiveweighting method and synthesized weighting method. The RBF-ANN models coupledwith the above mentioned multi-objective methods, whose parameters were optimizedwere by adaptive genetic algorithm, were utilized to search the optimal proportion ofthe compounds combinations. The results showed that the genetic algorithmconverged well.3. NSGA-II for multi-objective optimizationFor the traditional multi-objective methods the usual way was to transform themulti-objective problem into the single-objective problem. Although the evolutionaryalgorithms were proposed more later, they have some particular characteristics, suchas randomness, applicability and parallelism, which guarantee their efficiency andaccuracy. The evolutionary algorithms had been attracted much widespread concern.In this chapter a novel multi-objective optimization method, called NSGA-II, wasexplored for the multi-objective optimization research and compared with the abovementioned traditional multi-objective methods. The RBF-ANN models we had built inpart one were applied for NSGA-II and a Pareto-optimal solutions set with fiftynon-inferior solutions were obtained. The optimal proportion of the compoundscombinations resulted in NSGA-II was compared with the ones in other traditionalmulti-objective methods.