Data mining-based statistical analysis of biological data uncovers hidden significance: clustering Hashimoto’s thyroiditis patients based on the response of their PBMC with IL-2 and IFN-γ secretion to stimulation with Hsp60

The pathogenesis of Hashimoto’s thyroiditis includes autoimmunity involving thyroid antigens, autoantibodies, and possibly cytokines. It is unclear what role plays Hsp60, but our recent data indicate that it may contribute to pathogenesis as an autoantigen. Its role in the induction of cytokine production, pro- or anti-inflammatory, was not elucidated, except that we found that peripheral blood mononucleated cells (PBMC) from patients or from healthy controls did not respond with cytokine production upon stimulation by Hsp60 in vitro with patterns that would differentiate patients from controls with statistical significance. This “negative” outcome appeared when the data were pooled and analyzed with conventional statistical methods. We re-analyzed our data with non-conventional statistical methods based on data mining using the classification and regression tree learning algorithm and clustering methodology. The results indicate that by focusing on IFN-γ and IL-2 levels before and after Hsp60 stimulation of PBMC in each patient, it is possible to differentiate patients from controls. A major general conclusion is that when trying to identify disease markers such as levels of cytokines and Hsp60, reference to standards obtained from pooled data from many patients may be misleading. The chosen biomarker, e.g., production of IFN-γ and IL-2 by PBMC upon stimulation with Hsp60, must be assessed before and after stimulation and the results compared within each patient and analyzed with conventional and data mining statistical methods.