We present a method that analyses and visualizes individuals' health status and their treatment response as derived from multiple, multi-omics parameters of predefined biological processes. Using this method, we showed that an anti-inflammatory dietary intervention lead to significant average changes in plasma parameters and that unique individual responses could also be identified. These inter-individual responses were clustered into two distinct subgroups.
The treatment response (i.e., the movement in the health space) did not depend on age, BMI, order of treatment, waist-hip ratio, pre-weight, post-weight or on bodyweight change. The sub-group analysis revealed that individual treatment responses differed from the average group response. For instance, the treatment effect on plasma indole-3-propionic acid concentration was different between the two subgroups. Indole-3-propionic acid is a metabolite that originates from intestinal microbiota . Changes in this metabolite suggest adjustments in size or composition of the intestinal flora after dietary-mix intervention in subjects from responder group 1. The data suggest that the changes in intestinal flora occurred in subjects in group 1, but not in group 2.
The health space also reveals some personalized responses to the dietary intervention. Subject 14, for instance, was already in the improved health group when taking placebo, as a result of relatively high concentrations of fish-oil related parameters. This suggests that this person consumed more fish than average. Additionally, subject 1 responded differently than the other subjects. High plasma ALT and albumin concentrations indicated a biliary obstruction in this subject. A reduction in bile acid release may also reduce the absorption of fatty acids (and lipids and fat soluble vitamins) from the intestine and cause the observed lack in metabolic response. Moreover, this person had a high pro-oxidative-stress response which is known to be related to biliary obstruction [14, 15]. Also, the position of subject 19 in the health space was distinctly different than the others. Subject 19 was diagnosed with a malignant tumor (Kahler's disease) two months following the study. This disease may have caused the high levels of VCAM-1 in this subject. The increase of uric acid  and beta2-microglobulin  after the second period (for this subject the placebo period) indicate the onset of renal dysfunction during the study, which is a known complication of this disease.
The current version of the method builds PLS-DA models for the fold changes of the different parameters involved in the specific processes (scaled to 1). This method presumes that the importance of changes in plasma concentration is relative to the size of the change. This may not necessarily be correct. Certain minor plasma metabolite or protein changes may have large biological consequences, while other large changes may be of little biological relevance. This effect can be taken into account by including additional biological information, e.g. by weighting the components in the different axes. The current model includes only those parameters that significantly changed in the group average analysis. This may underestimate parameters that were affected on an individual basis, i.e. the metabolites and proteins that may be important for personalized health. A model variant may thus include all measured parameters relevant to the selected biological processes, instead of only the significantly changed processes. Another variant being developed includes multivariate feature selection by bootstrapping to select the most significant factors.
In this example of the 'health space' model, we used the processes metabolism, oxidation and inflammation. Some molecules are represented in more than one process (Additional file 1). This way we could take into account that the processes are biologically interdependent.
In the current example, 'health' was simply defined as the average state of metabolism, oxidation and inflammation after the anti- inflammatory intervention (the origin of the model). This demonstrates the applicability of the model. Of course, 'health' should be more accurately defined as a more absolute center (the origin) of the 'health space' model. Furthermore, genetic, life stage and environmental factors influence the chosen biological processes and have an impact on (optimal) health. Other biological processes may be taken into consideration. Multiple studies could be analyzed to more precisely define and refine the health space.