A Network-Constrain Weibull AFT Model for Biomarkers Discovery

We propose AFTNet, a novel network-constraint survival analysis method basedon the Weibull accelerated failure time (AFT) model solved by a penalizedlikelihood approach for variable selection and estimation. When using thelog-linear representation, the inference problem becomes a structured sparseregression problem for which we explicitly incorporate the correlation patternsamong predictors using a double penalty that promotes both sparsity andgrouping effect. Moreover, we establish the theoretical consistency for theAFTNet estimator and present an efficient iterative computational algorithmbased on the proximal gradient descent method. Finally, we evaluate AFTNetperformance both on synthetic and real data examples.