Research On Cancer Feature Gene Selection Based On Microarray Data

With the development of DNA microarray technology,the application of DNA microarray technology in gene diagnosis and auxiliary diseases is becoming more and more popular.Nowadays,in the case of high incidence of cancer lesions,DNA microarray technology is introduced to help humans explore information on biomolecules,it is a great significance in improving the success rate of cancer cure.However,the gene expression profile data derived from DNA microarray technology has the characteristics of highdimensional and small samples,and it is difficult to directly analyze the data.Therefore,the research on efficient feature selection algorithms has attracted the attention of scholars.According to the characteristics of gene expression profile data,two feature gene selection algorithms were proposed to improve the accuracy of cancer gene classification.Based on the classical genetic algorithm,a cancer feature selection and classification method AGALA was proposed.It combine adaptive genetic algorithm and learning automaton.The method adjusts the crossover rate of the crossover operation and the mutation rate of the mutation operation according to the size of the individual fitness value,to balance the global search ability and the local search ability of the algorithm.At the same time,it adds the reward and punishment operation of the learning automata,enhances the ability of the algorithm to search for new individuals,avoids the phenomenon of “premature maturity” in the late iteration of the algorithm,and speeds up the algorithm to search for the optimal subset of feature genes.Based on the standard particle swarm optimization algorithm,an SRPSO algorithm with self-adaptive and reverse-learning mechanism is proposed.The SRPSO method first uses T-test to filter data to remove redundant genes and reduce the search burden of the algorithm.Then SRPSO is used as the search engine of feature space,and combined with support vector machine,selects the characteristic genes with strong classification performance.The SRPSO algorithm uses the adaptive population iteration number to adjust the learning factor to control the speed of the particle search optimal position,and uses the reverse learning mechanism to prompt the algorithm search for new individuals and prevent the algorithm from stagnation in the late iteration.Experiments results show that compared with the traditional algorithm,the two algorithms proposed in this paper have higher classification accuracy,and the size of the feature gene subset obtained on the cancer gene expression spectrum is smaller,which is beneficial to improve the accuracy of cancer classification.