| 要約: | As a result of the current proliferation of scientific data of unprecedented magnitude and
complexity, ultrahigh dimensional data has become recurrent in a multitude of biological
studies. With biomarker identification being a key concern for early disease detection, the
ultrahigh dimensionality of data further complicates the complexity of the problem. Feature
screening has become increasingly significant in many scientific research but very limited
studies consider two types of survival endpoints, consider gene-gene dependencies and ac-
count for outliers. In this paper, we enhance joint correlation rank (JCR) screening by
utilising Google’s PageRank matrix to incorporate covariate-covariate network information.
A nonparanormal approach was also adopted to enable the screening to be more robust to
outliers. Through a series of simulations, we highlight its improved performance on identi-
fying active covariates accurately. For illustration, the proposed method is applied to colon
cancer data, where it is assessed based on prediction performance.
|
|---|