The DIVISION OF BIOSTATISTICS

In recent years, high-throughput scientific studies without any clear hypothesis have become very common. For example, a large-scale genomic study of a disease may examine thousands of genes without hypothesizing that any specific gene that is responsible for the disease. In these studies, the objective is to explore a large number of possible factors (e.g., genes) in order to identify a small number of them for follow-up studies that tend to be more thorough with much smaller scales. For such studies, we propose a nonparametric Bayesian approach based on Dirichlet process mixtures. Our model divides the set of candidate factors (e.g., genes) into several subgroups according to their degrees of relevance, or potential effect, in relation to the outcome of interest (e.g., disease status). We assign a rank to each factor according to the overall relevance of its corresponding group. The posterior expectation of these ranks can be used to set up a threshold for selecting relevant factors. Using simulated data, we demonstrate that our approach could be quite effective in finding relevant genes compared to several alternative methods. We also apply our model to two large-scale studies, where the objective is to identify differentially expressed genes between two types of leukemia, and Human Cytomegalovirus (HCMV) Infection, respectively.