![]() The key aspects of these methods and their growing success and future challenges include: the use of genetic variation and/or drug perturbation combined with transcriptional or other molecular profiling changes and clinical outcomes to go beyond correlation to extract causal interactions the use of large-scale supercomputing involving thousands or tens of thousands of processors to score billions of possible network frag- ments and evaluate billions of system-wide network hypotheses that can best explain the data having a sufficient number of patients or biological samples of appropriate genetic diversity and/or the use of appropriate animal models as surrogates for human diseases. Groups such as ours have developed technology that formalizes these approaches and applies them to the discovery of biomarkers and drug targets. This work led to the discovery of several novel drug targets and drug candidates (E Schadt, personal communication). Bayesian network inference algorithms applied to these well- designed data sets revealed a complex interconnected circuitry of several hundred genes that drive the low density lipoprotein, high density lipoprotein, and triglyceride metabolism, and body mass index of these mice that was subsequently validated in a large human cohort. Here the genetic diversity acted as the ‘ perturbation ’ or ‘ driver ’ that allowed correlation between various components to be interpreted as ‘ causal ’ network interactions. Sieberts and Schadt pioneered the creation and application of methods that linked genetic alterations to transcriptional changes and physiological outcomes in a genetic cross of two inbred mouse strains, one with a propensity for obesity and diabetes, and the other with a genetic resistance to obesity and diabetes. But some exciting progress has been made. We are still far from having comparable blueprints for the pathophysiologies of disease: connections are still miss- ing, and systematic measurements to parameterize and otherwise validate models are still hard to make. Black box models also work, provided the inputs and outputs are robustly validated. Classical engineering solutions depend on a complete and accurate blueprint of the system. While many of these efforts are impressive and in some rare cases have the ability to make non-intuitive predictions supported by experiment, these approaches address only a tiny fraction of the total possible circuitry of human biology and disease. Two contrasting approaches for unraveling the behavior of underlying genetic circuits and biochemical pathways are: (a) the curation of well-known biological pathways from the literature and conversion to descriptive visual displays depicting them as maps to visualize molecular profile changes in their context, Ingenuity Systems, and GeneGo Metacore pathway analysis] and (b) the construction of systems of differential equations that model the time evolution of gene products and their connection to phenotypic changes. ![]() a number of different approaches to interpret- ing large sequencing datasets and, increasingly, these fall within the realm of systems or network biology. ![]()
0 Comments
Leave a Reply. |