The concentration of metabolic products (metabolites) in various body fluids depends on the genetic and environmental exposure of a human being. This individual metabolic pattern is subject to normal physiological variation, but may also reflect pathological processes in the body.
Nuclear magnetic resonance (NMR) spectroscopy is the method of choice to detect metabolic constellations, as NMR allows the simultaneous quantification of ~400 metabolites in human specimens.
Finding metabolic constellations among hundreds of signals is not easy. We use machine learning to analyze study data together with NMR metabolite measurements to identify the markers that carry the most information about the disease in question and to build an equation that represents the metabolic constellation. This equation is then used as a diagnostic test in our products.
Machine learning comprises a number of methods within the field of artificial intelligence. Basically, these are very sophisticated statistical algorithms that use labeled examples of both healthy and sick subjects to identify the signals that separate the two (“learning the difference”). The resulting models from this metabolic constellation can then be used to predict if a patient is healthy or sick.
Our highly skilled statisticians ensure that the machine learning algorithms are used expertly. We complement this with the knowledge of our biochemists to guide the learning process with human knowledge. Combining the strengths of human and machine we ensure best possible results and allow our models to be understood by human experts like physicians.