Pancreatic Cancer Biomarker Detection By Improved Support Vector Machine For Recursive Feature Elimination

Nowadays,cancer has developed into an important fuse that endangers people's health,and clinical cases are increasing year by year.In many types of cancer,some malignant tumors usually have poor prognosis,and pancreatic cancer can be said to be one of them.If no treatment is given to the confirmed pancreatic cancer patients,the survival time is only a few months.Even if the operation is performed,the survival time is only about one year.According to the American Association for cancer statistics in male and female cases,pancreatic cancer is now among the top 10 cancers with high mortality.However,up to now,there has not been a complete and reliable set of early detection of pancreatic cancer markers.Therefore,in order to reduce the death rate of pancreatic cancer,early screening and timely treatment is particularly important.With the rise of machine learning,there are more and more ways to integrate machine learning algorithm into various researches.Based on machine learning method,this paper proposes a method to explore the clinical markers of pancreatic cancer by combining SVM-RFE and LDM-RFE.The cancer experimental data used in this paper are all from the geo database.In the preprocessing stage,we also preprocessed the data of ten other GSE datasets including breast cancer,gastric cancer,lung cancer,liver cancer,prostate cancer,colorectal cancer,etc.,and got ten feature lists respectively.Then we compared the features of the candidate feature list of pancreatic cancer with the above list one by one,and deleted the same feature genes in the feature list of pancreatic cancer.Through the above methods,730 pancreatic cancer specific differentially expressed genes were screened.By combining the recursive feature extraction method based on support vector machine and the recursive feature extraction method based on large interval distribution machine,we get a set of stable ranking list of specific candidate features.Finally,by comparing our method with those of random method,t-test method,SVM-RFE method and LDM-RFE method,we found that the combined classification effect was the best,among which the top seven differential expression genes(MMP7,mmp12,anpep,FOS,SFN,IL6 and A2M)were predicted to be the specificity of pancreatic cancer Biomarkers,becausethey have the best classification results for cancer and normal samples.After consulting the literature,we found that the abnormal expression of these genes was related to some diseases.In addition,through the further analysis of R2 platform(genome analysis and visualization platform),we once again proved that the seven genes selected above are closely related to the survival rate of pancreatic cancer patients.In the process of urinary excretion protein analysis,we found that the protein encoded by three genes(MMP7,FOS and A2M)can be secreted into urine to become urinary excretion protein,which helps to become the potential basis for clinical screening of pancreatic cancer.