LAGO: A Computationally Efficient Approach for Statistical Detection

• Activity Number: 377
• Type: Invited
• Date/Time: Wednesday, August 10, 2005 : 10:30 AM to 12:20 PM
• Sponsor: SSC
• Author(s): Mu Zhu*+
• Companies: University of Waterloo
• Keywords: Average precision ; Drug discovery ; Kernel density estimator ; Nearest neighbor ; Radial basis function (RBF) network ; Support vector machine (SVM)

Abstract

In drug discovery, chemists are faced with the problem of having to screen a large pool of candidates and identify a small number of chemical compounds worthy of further investigation. The drug discovery problem is an example of a more general class of statistical detection problems where the underlying objective is to detect items belonging to a rare class from a large database. We propose a computationally efficient method to achieve this goal. Our method consists of two steps. First, we estimate the density function of the rare class alone with an adaptive bandwidth kernel density estimator. The adaptive choice of the bandwidth is inspired by the ancient Chinese board game known as "Go." We then adjust this density locally depending on the density of the background class nearby. We show the amount of the adjustments needed are approximately equal to the adaptive bandwidths of our choice, which gives additional computational savings. We name the resulting method LAGO for "locally adjusted Go-kernel" density estimator and apply LAGO to a real drug discovery dataset, comparing its performance with a number of existing methods.