Cancer prognosis prediction using somatic point mutation and copy number variation data: a comparison of gene-level and pathway-based models

摘要

Workflow of gene-level models. In the 5 boxes are 5 intersected sets we achieved after filtering. “All” represents all the genes in the data, “Path” represents all the genes in the pathway collection, “COSM” represents all the genes in the COSMIC database, “Cox” represents all the genes which have significant p value in the univariable Cox model. The gene subsets represented by the 2 boxes after the dashed arrows were only applied to SPM data. 5-fold cross validation is conducted and for each training set, Cox Lasso is fitted. The estimated Cox model is applied to the test set to assess predictive performance. Unpenalized Cox models are also fitted, with the average value of lambda as the regularization parameter, to obtain non-shrunken coefficient estimates and p values