抄録
Relating gene expression profiles from microarray experiments with biological knowledge databases is an important step for understanding biological mechanism. It motivates the development of statistical techniques that quantify the significance of the expression profiles for sets of genes defined, e.g., based on pathway information. In this report we propose a new approach for gene set significance analysis. The proposed test measures the significance of gene sets using out-of-sample performance of nearest-neighbor classifiers. Our approach is computationally efficient and powerful to detect various types of differences in expression patterns. We demonstrate the advantage of our approach through the simulation studies for power analysis and an application to a public microarray data.
