Nonparametric Subcluster Detection in Large Hyperspaces.

Abstract

This assessment of subcluster detection in analytical chemistry offers a nonparametric approach to address the challenges of identifying specific substances (molecules or mixtures) in large hyperspaces. The paper introduces the concept of subcluster detection, which involves identifying specific substances within a larger cluster of similar samples. The BEST (Bootstrap Error-adjusted Single-sample Technique) metric is introduced as a more accurate and precise method for discriminating between similar samples compared to the MD (Mahalanobis distance) metric. The paper also discusses the challenges of subcluster detection in large hyperspaces, such as the curse of dimensionality and the need for nonparametric methods. The proposed nonparametric approach involves using a kernel density estimator to determine the probability density function of the data and then using a quantile-quantile algorithm to identify subclusters. The paper provides examples of how this approach can be used to analyze small changes in the near-infrared spectra of drug samples and identifies the benefits of this approach, such as improved accuracy and precision.