Characterization of Metabolic Interrelationships and in Silico Phenotyping of Lipoprotein Particles Using Self-organizing Maps

Plasma lipid concentrations cannot properly account for the complex interactions prevailing in lipoprotein (patho)physiology. Sequential ultracentrifugation (UCF) is the gold standard for physical lipoprotein isolations allowing for subsequent analyses of the molecular composition of the particles. Due to labor and cost issues, however, the UCF-based isolations are usually done only for VLDL, LDL, and HDL fractions; sometimes with the addition of intermediate density lipoprotein (IDL) particles and the fractionation of HDL into HDL(2) and HDL(3) (as done here; n = 302). We demonstrate via these data, with the lipoprotein lipid concentration and composition information combined, that the self-organizing map (SOM) analysis reveals a novel data-driven in silico phenotyping of lipoprotein metabolism beyond the experimentally available classifications. The SOM-based findings are biologically consistent with several well-known metabolic characteristics and also explain some apparent contradictions. The novelty is the inherent emergence of complex lipoprotein associations; e.g., the metabolic subgrouping of the associations between plasma LDL cholesterol concentrations and the structural subtypes of LDL particles. Importantly, lipoprotein concentrations cannot pinpoint lipoprotein phenotypes. It would generally be beneficial to computationally enhance the UCF-based lipoprotein data as illustrated here. Particularly, the compositional variations within the lipoprotein particles appear to be a fundamental issue with metabolic and clinical corollaries.