Comparison and Evaluation of Pathway and Gene-level Methods for Cancer Prognosis Prediction

摘要

Cancer prognosis prediction has become an important research goal. A limited number of gene-level analyses have led to clinically useful methods like Oncotype DX but these remain very difficult to develop and implement. A promising direction for improving the performance and interpretation of expression-based predictive models involves aggregating gene-level data into biological pathways. Although a few studies have used pathway-level predictors, a comprehensive comparison of pathway-level and gene-level prognostic models has not been performed. To address this gap, we characterized the performances of penalized Cox proportional hazard models built using either pathway or gene-level predictors for the cancers profiled in The Cancer Genome Atlas (TCGA) and pathways from the Molecular Signatures Database (MSigDB). When analyzing the TCGA data, we found that pathway-level models are more parsimonious and easier to interpret than the gene-level models without a loss of predictive performance. For example, both pathway and gene-level models have an average Cox concordance index of 0.85 for the TCGA glioma cohort, however, the gene-level model has twice as many predictors on average and the predictor composition is less stable across cross-validation evaluations. In simulations, when the correlation structure of the real data is broken, the pathway-level models have greater predictive performance and superior interpretative power relative to the gene-level models. For example, the average concordance index of the pathway-level model is 0.88 while the gene-level model falls to 0.56 for the TCGA glioma cohort.