Evaluation of a discrete dynamic systems approach for modeling the hierarchical relationship between genes, biochemistry, and disease susceptibility

A central goal of human genetics is the identification of combinations of DNA sequence variations that increase susceptibility to common, complex human diseases. Our ability to use genetic information to improve public health efforts to diagnose, prevent, and treat common human diseases will depend on our ability to understand the hierarchical relationship between complex biological systems at the genetic, cellular, biochemical, physiological, anatomical, and clinical endpoint levels. We have previously demonstrated that Petri nets are useful for building discrete dynamic systems models of biochemical networks that are consistent with nonlinear gene-gene interactions observed in epidemiological studies. Further, we have developed a machine learning approach that facilitates the automatic discovery of Petri net models thus eliminating the need for human-based trial and error approaches. In the present study, we evaluate this automated model discovery approach using four different nonlinear gene-gene interaction models. The results indicate that our model-building approach routinely identifies accurate Petri net models in a human-competitive manner. We anticipate that this general modeling strategy will be useful for generating hypotheses about the hierarchical relationship between genes, biochemistry, and measures of human health.