Scope and Method of Study: The ultimate goal of this research project was to complete the development of a simple, direct alternative method for the simultaneous quantitative determination of cholesterol and polyunsaturated fatty acids (PUFAs) in human serum by exploitation of various chemometric algorithms and consequent validation with the gas chromatography-mass spectrometry (GC-MS). In addition, oleic acid (OA) was also added as the eighth component and the performance of the various chemometric algorithms were compared. The study was also extended to various food and biological samples and chemometric algorithms were applied to obtain meaningful information of the data set.

Findings and Conclusions: For the first part of the study, ridge regression (RR), P-matrix (PM), principal component regression (PCR), and partial least squares (PLS2) algorithms performed quite equally well enough than the K-matrix (KM) approach when applied to the study of prepared mixtures (synthetic sera) in chloroform solutions. The PLS in the form of PLS2 model was tested for intact human serum specimens, and yielded results for w-3 and w-6 PUFA data that are comparable when using the GC-MS. Similar results were also derived for the between methods w-6/w-3 ratios. The first part of the study, therefore, showed the dominance of PLS2 over the other chemometric models.

The second part of the study showed that PLS1 algorithm yielded the least root mean square error of prediction (RMSEP) for all the lipid components as compared to all other algorithms. PLS1 yielded molar concentrations quite comparable with the GC-MS in the actual human serum samples. Inclusion of OA yielded high RMSEP despite attempts of utilizing the most robust algorithms like PLS1, PLS2, and PCR. GAPLS was able to successfully reduce the RMSEP for all the components over the non-GA PLS1 approach except for EPA and DHA. The spiking of human serum samples was also done in the study but the task is considered tedious for a typical clinical setting.

In a study involving OA, LA, and LNA in vegetable oils, it has been shown that PCR, PLS2, and PLS1 algorithms compared quite equally well in the prediction sets and that PLS2 mostly yielded a better performance than PLS1 and PCR algorithms in the unknown samples.

An extension of the assay was performed for the pattern recognition of biological and food samples. The assay was able to discriminate eleven clusters corresponding to different food and biological samples.

The study has shown how the Purdie assay coupled with chemometric algorithms might provide alternatives to separations methods for the direct determination of lipids in human serum, vegetable oils, and their synthetic models. The advantages of this simple technology are the reduction in time and costs.