Selection Of Polynomial Model And Design Of Experiment In Drug Combination Optimization

Complex systems are the frontier of current scientific researches.Organisms are complex systems composed of various molecules,which are closely related to the development of many diseases.Thus,combination drugs tend to be more effective and less toxic as they acting on multiple nodes in the complex systems.However,how to screen and optimize the combination drugs,that is,to establish the relationship between multiple inputs and the phenotypic output of complex systems,is a difficult problem and challenge in the field.In order to solve this problem,feedback system control(FSC)technology combines the model building ability of machine learning and the search ability of heuristic algorithms to iteratively summarize and compare a small amount of experimental data in a systematic way,so as to find the optimal drug combination.Nevertheless,there are still two deficiencies in the current FSC methods.First,the optimization efficiency of FSC depends on whether the model of the dose-response relationship is accurate,while there is currently no systematic comparison of existing model selection and building methods under different application conditions.Second is the lack of systematic guidance for optimization and experimental design strategies under different numbers of drugs.In response to these shortcomings,this thesis systematically summarizes the drug combination optimization based on computer simulation.From the perspective of complex systems,this paper theoretically demonstrates the feasibility of using quadratic polynomial modeling to optimize drug combinations,and compares the efficiency of single-iteration modeling and FSC based algorithms for drug combination optimization.Based on the simulation results,we devised a new stepwise regression algorithm using cross-validation,and combined it with FSC scheme.We have compared the optimization efficiency of the new algorithm with other FSC algorithms under different conditions.Finally,this paper also discusses the minimal amount of experiments required to optimize drug combinations via the FSC based algorithm under different conditions.