Ellipsoid clustering machine: a front line to aid in disease diagnosis

Abstract

This study presents a new machine learning strategy to address the disease diagnosis classification problem that comprises an unknown number of disease classes. This is exemplified by a software called Ellipsoid Clustering Machine (ECM) that identifies conserved regions in mass spectrometry proteomic profiles obtained from control subjects and uses these to estimate classification boundaries based on sample variance. The software can also be used for visual inspection of data reproducibility. ECM was evaluated using mass spectrometry protein profiles obtained from serum of Hodgkin’s disease patients (HD) and control subjects. According to the leave-one-out cross validation, ECM completely separated both groups based only on the information derived from four selected mass spectral peaks. Classification details and a 3D graphical model showing the separation between the control subject cluster and HD patients is also presented. The software is available on the project website together with online interactive models of the dataset and an animation demonstrating the method.