B2004
Title: Differential Projection pursuit methods and its applications to differential experiments
Authors: Javier Cabrera - Rutgers University (United States) [presenting]
Yajie Duan - Rutgers University (United States)
Abstract: The novel concept of differential projection pursuit, and its applications to the analysis of large datasets, are introduced. Projection pursuit has been applied for many years as a standard methodology for analyzing multivariate data. But in the applications of projection pursuit in the experimental setting, there are two issues of importance, which are the large number of observations and the differential nature of most experiments. The differential projection pursuit methodology objective is to find projections that maximize the difference between two or more treatments or distributions. We will introduce a new index, similar to the Natural Hermite index, that is suitable for measuring differences between 2 or more distributions. This implementation of Differential Projection Pursuit is also suitable for datasets with small and large numbers of observations, such as flow cytometry datasets. We will also present a differential projection pursuit analysis of a large flow cytometry dataset with a treatment sample and a control sample. The algorithm will search for optimal projections and display clusters of treated cells in regions where there are few control cells. A rotation will be applied to align the axes of the optimal projections with the original variables on the dataset, for better interpretation of the results.
