John T. Magee

On mass spectrometer instrument standardization and interlaboratory calibration transfer using neural networks

For pyrolysis mass spectrometry (PyMS) to be exploited in areas such as the routine identification of microorganisms, for quantifying determinands in biological and biotechnological systems, and in the production of useful mass spectral libraries, it is paramount that newly acquired spectra be comparable to those previously collected and held in a central reference laboratory. Artificial neural networks (ANNs) and other multivariate calibration models have been used to relate mass spectra to the biological features of interest. However, calibration models developed on one mass spectrometer cannot be used with spectra collected on a second instrument, because of the differences between the instrumental responses of both instruments. We report here that an ANN-based drift correction procedure can be implemented so that newly acquired spectra can be used to challenge models constructed using mass spectra collected on different instruments. Calibration samples were run on three different PyMS machines, and ANNs set up in which the inputs were the 150 machine ‘a’ calibration masses and the outputs were the 150 calibration masses from the machine ‘b’ spectra. Such associative neural networks could thus be used as signalprocessing elements to effect the transformation of data acquired on one machine to those which would have been acquired on a different instrument. Therefore, for the first time PyMS could be used to acquire spectra which could usefully be compared to those previously collected and held in a data-base, irrespective of the mass spectrometer used. The examples reported are for the quantitative assessment of the amount of lysozyme in a binary mixture with glycogen and the rapid identification down to the species level of bacteria belonging to the genus Eubacterium. This approach is not limited solely to pyrolysis mass spectrometry but is generally applicable to any analytical tool which is prone to deterioration in calibration transfer, such as IR, ESR, NMR and other vibrational spectroscopies, gas and liquid chromatography, as well as other types of mass spectrometry.

Detection of Infectious Outbreaks in Hospitals through Incremental Clustering

This paper highlights the shortcomings of current systems of nosocomial infection control and shows how techniques borrowed from statistics and Artificial Intelligence, in particular clustering, can be used effectively to enhance these systems beyond confirmation and into the more important realms of detection and prediction. A tool called HIC and examined in collaboration with the Cardiff Public Health Laboratory is presented. Preliminary experiments with the system demonstrate promise. In particular, the system was able to uncover a previously undiscovered cross-infection incident.