Human chromosome classification using discriminant analysis and Bayesian probability

Abstract

A computer-aided method of karyotyping human chromosomes was presented which utilized the probability of classification in a discriminant analysis as a measure of the confidence of correct identification. The program assigned each chromosome to one of two partial karyotypes based on a high or low probability of identification. A secondary discriminant analysis classification was then carried out on chromosomes identified with low confidence to determine whether additional chromosomes could then be reclassified with a higher level of confidence. The method was tested using 20 GTG-banded cells and 44 RFA-banded cells. The latter were divided into a 34-cell learning set and a 10-cell test set. Among the GTG-banded chromosomes used, 90.6% were classified with high confidence; of these, 93.5% were correct. A total of 88.6% were correct, however, if the confidence level was not considered. 77.3% of the RFA-banded learning set and 75.2% of the RFA-banded test set chromosomes used were classified with high confidence; 94.1% and 85.8%, respectively, were correct. If the confidence level was not considered, 86.1% of the learning set and 79.7% of the test set chromosomes were correctly identified.