Pattern Recognition for Discrimination of Dyslipidemic States

The Centers for Disease Control and Prevention reported that the diseases of the heart, cancer, stroke, Alzheimer's disease, and diabetes are among the top leading causes of deaths in the US for 2007. These diseases are known to be caused by a variety of factors including cholesterol and polyunsaturated fatty acids (PUFAs). This paper shows how pattern recognition using an unsupervised clustering algorithm (principal component analysis (PCA)) can provide a direct method of discriminating dyslipidemic patients according to Fredrickson's Classification of Dyslipidemias obtained from the spectral data. The spectral data were obtained by utilizing a mature, patented reagent system based on its selectivity to the -CH=CH-CH2- group in a wide variety of lipids in human serum samples. The simple colorimetric assay used is rapid, rugged, and inexpensive that produces a characteristic molar absorbance spectra for cholesterol, omega-3 (methyl esters of linolenic, eicosapentaenoic acid (EPA), and docosahexaenoic acid (DHA) fatty acids), and omega-6 (methyl esters of linoleic, conjugated linoleic acid (CLA), and arachidonic fatty acids). The assay is reacted with serum samples of known dyslipidemias. An independent full factorial design simulation of synthetic serum mixtures containing different levels of the lipids in chloroform solutions was also done. Pattern recognition using an unsupervised clustering algorithm (PCA) was applied to both the spectral data of synthetic and actual serum samples. The full factorial design of synthetic mixtures of human serum in chloroform solutions accomplished the discrimination of eight clusters. Each cluster corresponded to specific levels of lipids prepared. Application of the assay to real serum samples revealed ten clusters with each corresponding to a disease state according to Fredrickson's Classification of Dyslipidemias. The results demonstrated that the novel chemical approach could provide a good agreement between clinical chemistry and pattern recognition results. This study shows how pattern recognition coupled with the assay can provide discrimination of dyslipi-demic states in dyslipidemic individuals in faster and cheaper ways.