392
to achieve two new principal component axes which separated four of the five particle types
in the sample.
Clustering, the other major multivariate analysis technique, seems a logical choice for flow
cytometry, since it attempts to divide a set of multivariate observations into similar groups or
"clusters." As with ordination techniques, there are many different methods of clustering.
Non-hierarchical methods are probably best suited to the large data sets typical of flow
cytometry, since they optimize within-group homogeneity and do not require the entire data
set to be loaded into computer memory for analysis (Legendre and Legendre 1983).
A promising new technique for flow cytometric analyses is the use of neural net computer
systems (Frankel et al. 1989, Frankel and Vogt 1990). These computer systems are trained
with a known data set, as in discriminant analysis, and can then classify unknown samples
quickly and effectively. Neural nets require no mathematical model of the phenomena being
studied and make no assumptions about the form or distribution of the data. They can also
adapt to non-linear variations such as the increase in phytoplankton cell pigment content with
depth, something commonly used multivariate techniques do not handle well. A neural net can
be trained to take factors such as depth, season, time of day, and geographical location as well
as optical properties into account when classifying particles. Frankel et al. (1989) used
software versions and achieved near-real-time rates of analysis of marine field samples, so
faster hardware implemented versions should be able to process data as it is collected. A
similar approach has been used successfully to analyze mixtures of phytoplankton using
6-parameter data from the Optical Plankton Analyzer (Balfoort et al. 1990).
CONCLUSION
The study of individual particles in the sea using flow cytometry has changed the way we look
at the plankton; this new technology has opened our eyes to organisms and to relationships
in the plankton community that were not apparent from traditional bulk analyses. As advances
in molecular and cell biology allow us to probe the identity and physiology of individual
organisms with increasing specificity, we must develop the capacity to apply these advances
to achieve two new principal component axes which separated four of the five particle types
in the sample.
Clustering, the other major multivariate analysis technique, seems a logical choice for flow
cytometry, since it attempts to divide a set of multivariate observations into similar groups or
"clusters." As with ordination techniques, there are many different methods of clustering.
Non-hierarchical methods are probably best suited to the large data sets typical of flow
cytometry, since they optimize within-group homogeneity and do not require the entire data
set to be loaded into computer memory for analysis (Legendre and Legendre 1983).
A promising new technique for flow cytometric analyses is the use of neural net computer
systems (Frankel et al. 1989, Frankel and Vogt 1990). These computer systems are trained
with a known data set, as in discriminant analysis, and can then classify unknown samples
quickly and effectively. Neural nets require no mathematical model of the phenomena being
studied and make no assumptions about the form or distribution of the data. They can also
adapt to non-linear variations such as the increase in phytoplankton cell pigment content with
depth, something commonly used multivariate techniques do not handle well. A neural net can
be trained to take factors such as depth, season, time of day, and geographical location as well
as optical properties into account when classifying particles. Frankel et al. (1989) used
software versions and achieved near-real-time rates of analysis of marine field samples, so
faster hardware implemented versions should be able to process data as it is collected. A
similar approach has been used successfully to analyze mixtures of phytoplankton using
6-parameter data from the Optical Plankton Analyzer (Balfoort et al. 1990).
CONCLUSION
The study of individual particles in the sea using flow cytometry has changed the way we look
at the plankton; this new technology has opened our eyes to organisms and to relationships
in the plankton community that were not apparent from traditional bulk analyses. As advances
in molecular and cell biology allow us to probe the identity and physiology of individual
organisms with increasing specificity, we must develop the capacity to apply these advances