| Feature selection is one of the key problems in machine learning and data mining.It can reduce the redundancy of features and improve the classification accuracy at the same time.Feature selection plays a crucial role in effectively avoiding "dimension disaster" and in identifying biomarkers for cancer diagnosis.First,this paper propose an algorithm to generate Helper Gene.The algorithm also provides a transgenic island process,which is applied to feature selection.We also evaluated the number of helper genes added.In order to verify the performance of the helper genes in transgenic island algorithm,it is applied to the diagnosis of cancer.Gastric cancer is a common cancer type and prevalent in the eastern Asian population.Most gastric cancer biomarkers were detected by their top-ranked differential expressions between gastric cancers and controls.This paper hypothesize that the lowly-ranked features carry their unique contributions to the prediction task of gastric cancers.The proposed algorithm was used to find 43 lowly-ranked features whose integrations can improve the prediction performances of the top-ranked features selected by six conventional and three swarm intelligence feature selection(FS)algorithms.These 43 lowly-ranked features are called the helper genes,and their contributions to the nine FS algorithms were further confirmed on two independent gastric cancer datasets.This paper further observed that the nine FS algorithms were improved by the integrations of the 43 helper genes on two lung cancer datasets.This proof-of-principle study suggests the existence of the helper genes among the lowlyranked features that are ignored by many existing studies.And most biomarker studies may evaluate whether their disease prediction models may be improved by the integration of helper genes.In addition,Swarm intelligence(SI)algorithms are now widely used.Since it was proposed,Grey Wolf algorithm(GWO)has solved many problems,which is novel and valuable.An improved transgenic Grey Wolf Optimizer(ITGWO)algorithm is proposed.The combination of Grey Wolf Optimizer Algorithm and biological knowledge is realized for the first time.To be specific,during initialization,chaotic mapping is adopted to change the original random way,which leads to the phenomenon that the process of generating the optimal solution is very slow.In the stage of mutation,to enhance the development capability,a two-stage mutation algorithm was added.Through the first mutation,feature deletion is carried out under the condition that the classification accuracy will not be reduced.The second mutation pays more attention to the feature information,so the feature is added.Both mutations use a greedy mechanism to ensure that they do not mutate to a worse solution.Transgenic island algorithm is added.This algorithm is inspired by genetic modification,and in order to improve classification accuracy and reduce fitness values,it retains excellent features in the process of mutation.The proposed algorithm is compared with 10 algorithms such as TMGWO on 20 datasets,which fully demonstrates the superiority of the proposed algorithm in feature selection,especially in binary classification.Transgenic GWO provides a new idea for the application of swarm intelligence algorithm in feature selection. |
