Gene selection in Cox regression model based on a new adaptive penalized method

Abstract

<jats:p>The common issues of high dimensional gene expression data for survival analysis are that many of genes may not be relevant to their diseases. Gene selection has been proved to be an effective way to improve the result of many methods. The Cox proportional hazards regression model is the most popular model in regression analysis for censored survival data. In this paper, an adaptive penalized Cox proportional hazards regression model is proposed, with the aim of identification relevant genes and provides high classification accuracy, by combining the Cox proportional hazards regression model with the weighted least absolute shrinkage and selection operator (LASSO) method. Experimental results show that the proposed method significantly outperforms two competitor methods in terms of the area under the curve and the number of the selected genes. </jats:p>

Detecting prognostic biomarkers of breast cancer by regularized Cox proportional hazards models

“…Later, Li and Luan [ 22 ] proposed the CPH model with ridge penalty and clarified the limitation of using all genes for prediction, but it does not provide a method for selecting feature genes for prediction. In addition to the well-known norm (Lasso) and norm (Ridge), convex penalty functions such as the linear combination of and norm (i.e., Elastic net, abbreviated as Enet) have also been proposed for feature selection as well as model prediction [ 23 ]. Next, various other non-convex penalty functions, e.g., (the regularized representative of ) [ 5 ], [ 24 ], SCAD [ 25 ] and MCP [ 26 ], have good performance in sparse optimization.…”