A new algorithm for clustering lymphocyte typing sera.

An algorithm is presented for clustering antisera by computer. It has two novel features: the leading serum to which all other sera in the cluster are compared is chosen as the most centrally located serum in the cluster; the similarity between two sera is defined from the 2 X 2 table of serum reactions as s = 2a/(2a + b + c). This similarity index is a better measure of the similarity between two sera than conventional measures of similarity such as the correlation coefficient. Finally, the identification of cluster and serum subsets provides a more complete analysis of cross-reactivity and multispecificity, and suggests which absorptions might yield monospecific typing sera. A computer program which performs this serum cluster analysis is available upon request.