Additive Cox proportional hazards models for next-generation sequencing data

摘要: Eighty-Nine Non-Small Cell Lung Cancer (NSCLC) patients experience chromosomal rearrangements called Copy Number Alteration (CNA), where the cells have abnormal number of copies in one or more regions their genome, this genetic alteration are known to drive cancer development. An important aim thesis is to propose a way combine clinical covariate as fixed predictors with CNAs genomics windows smoothing terms using penalized additive Cox Proportional Hazards (PH) model. Most proposed prediction methods assume linearity genomic windows along covariates. However, continuous covariates can affect hazard via complicated nonlinear functional forms. Therefore, PH model are likely misspecified, because it is not fitting correct form for Some reports work on combining high-dimensional data based standard them focus applying variable selection CNA data. Our main interest procedure select effects from genomic-windows. Two different approaches feature selection presented which discrete and shrinkage. Discrete feature univariate selection, identify subset the CNAs genomic-windows have strongest survival time, while by shrinkage works adding second penalty partial log-likelihood, that leads penalizing coefficients model, result some coefficient being set zero. For NSCLC dataset, we find size tumor spread into lymph nodes significant factors increase patients survival, estimate smooth log ratio curves of the contribute higher lower death across genome.