The distorted shell method for clustering for syndrome classification.

Abstract

Syndrome classification may be described as the arrangement of individuals into groups on the basis of their phenotypic resemblance. This paper describes how phenotypic resemblance may be quantified and demonstrates a numerical method called distorted shell clustering, which isolates groups of phenotypically similar individuals representing syndromes. This new method takes into consideration apparent biological properties of syndromes. It allows for overlapping phenotypes between syndromes, and differing character association and variability within syndromes. This method is compared to four other clustering methods by using suspects for a syndrome of known etiology (Down syndrome). The numerical results based on the phenotype then can be compared with the actual diagnosis. Only the distorted shell method classifies patients, without error, into two major clusters: the Down and the non-Down, while maintaining a high level of efficiency.