The Research Of Correlation Filter In Tumor Subtype Classification

Cancer is a complex disease.It is the key that how to detect and treat it early to cure cancer.However,in terms of gene expression profiles,many tumors have different subtypes.Different tumor subtypes have different gene expression characteristics,so it is difficult to distinguish tumor subtypes according to clinical features.The proof of Gene expression profile can provide location information in any given cell under the condition which existing thousands of gene expression,thereby easily understanding tumor generation and development mechanism.It can accurately distinguish tumor subtypes in gene expression profile data.The data of gene expression profiles are characterized by high dimensionality,small sample size,high noise and high correlation.How to extract valid data from massive gene expression profiles to classify and identify certain tumor subtypes has become a challenging task.Based on the data of gene expression profiles,the classification of tumor subtypes is studied from the level of gene expression.The main research works are summarized as follows:(1)An improved optimal discriminate function filter method IOTSDF is proposed.There is a certain correlation between the gene expression profiles.The correlation pattern recognition method can construct the correlation filter to match the differentially expressed genes with the tumor subtypes determined by tumor related genes,so as to effectively avoid the problems caused by data correlation.By optimizing the control parameters of the filter,the IOTSDF filter has the characteristics of high noise tolerance and strong classification ability.Experimental results in real tumor datasets show that the IOTSDF filter method has high classification capability and can effectively suppress complex noise.(2)Based on the correlation filter,the maximum margin correlation filter MMCF is applied to tumor subtypes classification.Previous studies have shown that support vector machine SVM has better generalization performance in classification,and the correlation filter has the characteristic of high classification ability.MMCF filter combine the advantages of the SVM classifier with the correlation filter,which made the filter has better generalization ability and robustness in tumor subtype classification.The experiments on different gene expression datasets show that MMCF classifier has good classification performance and generalization performance.