Font Size: a A A

Quantitative Structure-activity Relationship Based On Improved Artificial Bee Colony Algorithm

Posted on:2021-04-21Degree:MasterType:Thesis
Country:ChinaCandidate:Y H LinFull Text:PDF
GTID:2428330614954794Subject:Statistics
Abstract/Summary:PDF Full Text Request
Quantitative Structure-Activity Relationship(QSAR)refers to using structural parameters(including physical and chemical parameters measured experimentally and theoretically calculated parameters)to describe the structure of compounds,and then to describe the interrelationship between the structure and activity of organic compounds.This method has been widely used in the fields of drug design and life sciences.The prediction of QSAR based on these parameters which are irrelevant to the activity of organic compounds or redundant is usually not accurate.Therefore,it is necessary to filter these parameters(features)to remove redundant or irrelevant parameters(features)to improve the accuracy of QSAR modeling.Artificial Bee Colony Algorithm(ABC)has strong global search ability,but it still has some shortcomings in feature selection.In this study,an improved ABC algorithm is proposed to solve the feature selection problem in QSAR modeling.The main research contents are as follows:First,the standard ABC algorithm is mapped from continuous space to discrete space for feature selection,and compared with other intelligent calculation methods for QSAR in feature selection.Experimental results show that: in QSAR,the feature selection performance based on the standard ABC algorithm is superior to other algorithms on the Artemisinin dataset;It is worse than other algorithms on the BZR and Selwood datasets.This shows that although the ABC algorithm can be used to solve the problem of feature selection in the process of QSAR modeling,there is room for improvement.Second,the essence of the standard ABC algorithm is still the optimization of continuous space.Algorithm 1 introduces crossover operators and mutation operators to achieve feature selection in discrete spaces.On this basis,the particle update method at the stage of employed bees and onlooker bees has been improved,and each particle in the population is cross-operated with randomly selected particles and mutation operations are performed to generate new particles.Algorithm 2 adds a feedback mechanism on the basis of Algorithm 1,so that each particle generated in the stage of scout bees learns from the optimal particle in the population.The experimental results show that,in all data sets,Algorithm 1 is superior to five algorithms such as Algorithm 2 onand root mean square error.Its value is also significantly better than other algorithms,indicating that Algorithm 1 can obtain very good prediction accuracy.Algorithm 2 is better than Algorithm 1 on the Artemisinin and Selwood datasets,but it's not better than all other algorithms on and root mean square error,indicating that the feedback mechanism plays a certain role in reducing the number of features,but reduces the prediction accuracy.Last,for the problem that standard ABC,algorithm 1 and algorithm 2 do not considervalue and feature number at the same time in the process of feature selection,an improved feature selection algorithm for multi-objective ABC is proposed.The improvement strategy is: select the optimal particle set of the Pareto front through non-dominated sorting,and extract the particle from the population at the employed bees stage and onlooker bees stage,and randomly extract the particles in the optimal particle set,and then the two are cross-mutated to generate 2 * N new particles.Experimental results show that the MOABC-PLS-4 algorithm obtains less features when the value remains relatively stable,and reduces the root mean square error to a certain extent.
Keywords/Search Tags:Quantitative structure-activity relationship, Artificial Bee Colony optimization algorithm, feature selection, feedback mechanism, multi-objective Artificial Bee Colony algorithm
PDF Full Text Request
Related items