Using the Precision Lasso for gene selection in diffuse large B-cell lymphoma cancer

Background and aims: Today, with the advancement of technology, the issues of high-dimensionaldata in various area of science are extensively discussed. This type of data has unusual andunstructured dimensions. Genetics is one area that deals with this type of data today. One of thegoals of genetic science is to provide appropriate tools for diagnosing diseases such as cancer,predicting disease and also responding to treatment based on gene expression profiles.Gene expression data of patients with diffuse large B-cell lymphoma are used based on microarraytechnology. In this type of high-dimensional data sets, the problem of high correlation betweenvariables is also discussed.The aim of this study is to perform Precision Lasso regression model on gene expression of diffuselarge B-cell lymphoma patients and to find marker genes related to DLBCL.Method: In the present case-control study, dataset has been used including, ۱۸۰ gene expressionfrom ۱۴ healthy individuals and ۱۷ DLBCL patients. The marker genes are selected by fittingridge, Lasso, Elastic Net, and Precision Lasso regression models. In addition, the predictive accuracyof each model will be examined by using the mean squared error. Finally, the best model isselected for diagnosing, predicting cancers.Results: Based on our findings, the Precision Lasso, the Ridge, the Elastic Net, and the Lassomodels choose the most marker genes, respectively. In addition, the top ۲۰ genes based on modelscompared with the results of clinical studies. The Precision Lasso and the Ridge models selectedthe most common genes with the clinical results, respectively. In order to evaluate the goodnessof fit of regression models with the mean squared error index, the Performance of the elastic netand ridge and Precision Lasso models is very suitable.Conclusion: In particular, these regression models are suitable for such dataset, including thenumber of explanatory variables greater than the number of observations, with a high correlationbetween variables. These models selected genes related to DLBCL cancer. The results were reportedby statistical and clinical comparison. performance of Precision Lasso model in selectingrelated genes could be considered more acceptable rather than other models.

کلیدواژه ها:

gene expression ، high-dimensional ، Precision Lasso ، lymphoma cancer