The advent of microfluidic technologies has raised the possibility of rapid, portable bioagent detection devices based on capillary gel and zone electrophoresis separations. However, these technological advances come with new challenges surrounding data interpretation. Specifically, the interpretation of complex detector signals "typically a time sequence of peaks that correspond to proteins in the sample" coupled with significant noise and uncertainty, makes it difficult to robustly determine the presence of a specific organism.

As part of Sandia's µChemLab™ project, CRF researchers are developing stochastic algorithms for robust inference of bioagent detection. Funded by the Department of Defense and the Department of Homeland Security, the µChemLab project seeks to develop fully self-contained, portable, hand-held chemical analysis systems incorporating "lab on a chip" technologies. A significant effort within this project is devoted to the development of detection signatures for providing robust inference of bioagents.

This effort relies on Bayesian inference techniques to provide a probabilistic detection based on measured data. Bayesian techniques offer advantages in terms of machine learning capabilities, pattern recognition, and classification in noisy environments. In this context, they provide rational means of combining prior information with observed data and arriving at a probabilistic statement regarding the presence of a specific organism in a sample. Such probabilistic inference is highly useful for decision making subsequent to detection. New algorithms and software are being developed for Bayesian classification among a range of potential bioagents. The resulting construction involves an initial training phase for teachingthe instrument the expected signatures of potential bioagents. These signatures will then be used for classification during active detection.