Method Research On Classification Of Multi-model Biometrics Based On Coupling Metric

Gene expression data are unique,difficult to copy and also be considered as a biometric.The classification of cancer molecular subtypes combined with the biological characteristics and clinical manifestations of cancer patients is of great significance in guiding the diagnosis and treatment of cancer.Most of the published research on the molecular subtype of cancer is based on a single genome data,especially the transcriptome data.Considering the heterogeneity of different levels of gene regulation,only use the transcriptome data does not necessarily observe complete biological characteristics.And the classifier of cancer molecular subtypes is only applicable to transcriptome data,and for other omic data such as miRNA data,it is not possible to use molecular-type information.Therefore,this paper attempts to apply the improved canonical correlation analysis(CCA)method to the bioinformatics,to realize the fusion of multi-modal data,and to establish a multi-model classifier.Finally,we use the labeled colorectal cancer data: mRNA and miRNA data for validation.The main research work of this paper is as follows:1.This paper first introduces the harmfulness of cancer,as well as some problems of the present stage of cancer treatment.Based on the cause and the overall description of the current global cancer,this paper shows the importance of precision medicine.Next,we introduce the current status of cancer research based on genome data,discuss the relationship between cancer molecular typing and clinical precise treatment,and elaborate the necessity of cancer classification based on multi-modal data.2.This paper briefly describes the establishment of cancer gene dataset for method validation,mainly including the acquisition of gene data,the selection of data based on expression variation,the standardization method of data.Then,this paper introduces the classifier and Wilcoxon ranked test used in this paper.3.This paper introduces the traditional correlation analysis algorithm(CCA),the improved CCA algorithm and the feature fusion method used in the next experiments.A comparative experiment was conducted to select the best feature fusion method and classifier combination.The multi-modal classifier was used to classify the multi-modal data and the single modal data respectively.4.From the perspective of clustering,this paper first labels colorectal cancer samples from the data point of view.Then establishing the classifier by multi-modal data and feature fusion.And compares the recognition rate of the cluster label and the cancer molecular label by leave-one-out method.From the perspective of cancer genome data,the idea of fusing multi-modal data with the improved canonical correlation algorithm provides a new idea for comprehensive utilization of cancer genomics data.In this paper,we have selected the most suitable methods of data preprocessing,feature fusion and designed classifiers through several groups of experiments.In view of the current situation of cancer research,cancer molecular classifier is only applicable to transcriptome data.The molti-modal classifier can achieve classification of other genomics data to some extent.It is of great significance to further study cancer by comprehensively using multi-modal information.