Statistical Design and Data Mining for Microsensor Arrays

• Activity Number: 458
• Type: Contributed
• Date/Time: Wednesday, August 10, 2005 : 2:00 PM to 3:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Nell Sedransk*+ and Zhan-Qian J. Lu
• Companies: National Institute of Standards and Technology and National Institute of Standards and Technology
• Keywords: experimental design ; high-dimensional data ; neural networks ; multivariate models ; data mining ; pattern matching

Abstract

Data from microsensor arrays designed with combinatorial and high-throughput measurement capabilities take the form of multiple simultaneous spectra. Statistical challenges involve design of experiments for analysis of high-throughput data under multifactor external influences, development of statistical methodology for rapidly extracting relevant signals from high-interference data (truly a "finding the needle in a haystack" problem), and characterization and comparison of the performance of different data mining tools applied to these high-dimensional signals. Ongoing multidisciplinary research at NIST uses microarrays of four or more sensors designed to allow simultaneous, synchronized monitoring on different film materials across a wide temperature range. Working jointly, statisticians and chemists have developed statistical designs and data mining tools that allow this low-cost and extra-sensitive technology to be used for detection of trace contaminants at ppb concentrations in the presence of strong interference. Statistical principles for design and performance of data mining tools are implemented and evaluated in the context of this research.