A Comparison of One- and Two-Dimensional Classification of Hemagglutination Reactions in Microplates

The ability of a two-dimensional classification system to improve detection of weak hemagglutination reactions was investigated. Two canonical variables were derived by stepwise linear discriminant techniques from a set of 24 measurements of light absorbance for each of 605 hemagglutination reactions. A test set of 3864 reactions were then classified by the k nearest neighbor algorithm using either the first canonical variable or both variables. No improvement in the ability to distinguish positive reactions from negative ones was found when the second canonical variable was added. This implies that all of the information about hemagglutination reactions in microplates is contained in the first canonical variable. Thus, computationally expensive two-dimensional pattern classifiers will not improve the performance of automated microplate hemagglutination detection. It appears that it will be necessary to reduce the size of the light beam used to measure light absorbance in order to improve detection of weak reactions.