Pareto Optimal Ensemble Techniques in Systems Biology
Biological systems are a complex network of myriad biomolecular species with thousands of interactions. In order to analyze key properties of this vast web, various techniques have been developed under the discipline of systems biology. These techniques simplify the biological networks, modeling key proteins as nodes and their major interactions as edges. For instance, a prototypical-signaling pathway for pain in sensory neurons is studied by Cornell PhD student Anirikh Chakrabarti in the Varner Lab. The model describes 90 species connected by 162 interactions. In order to quantify these interactions, flux balance equations are constructed and parameters, like reaction rate constants and initial conditions, are randomly generated. The model is then trained with experimental data to produce numerous parameter ensembles that provide global fit to the empirical results.
In order to select a suitable ensemble, one common approach is the use of Pareto Optimal Ensemble Techniques (POETs). In this study, the POETs algorithm integrates Simulated Annealing and multi-objective optimization through the notion of Pareto rank to optimally balance trade-offs. To illustrate this concept, consider a parameter ensemble with two sets of experimental data. The parameters are adjusted continuously to achieve better fits by minimizing the sum of differences. However, when two empirical data sets contradict each other, the parameters remain robust until a new parameter with a decreased error for at least one of the data set is found. After successive iterations, the absolute error for each ensemble is calculated and compared. The parameter ensemble with minimum error is then selected and further tested by sensitivity analysis. For details about the POETs algorithm and sensitivity analysis, please refer to the two attached articles or visit www.varnerlab.org for more examples of systems biology networks.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2735677/?tool=pubmed
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3021968/
wow this is very interesting. i learned so much