Abstract:
|
We consider time course gene expression profiles and the task of identifying genes that are differentially expressed across different groups (such as good/poor responser to a specific therapy), e.g., as a preselection step when building a predictive model. For doing so, we employ a nonparametric technique: a functional nearest neighbor ensemble with implicit variable selection. We evaluate the performance of the new method in simulation studies and on real world data, and compare it to alternative procedures such as gene-wise testing.
|