A network approach for the mechanistic classification of bioactive compounds

• Main Author: Siebert, Trina A.
• Other Authors: Chaplen, Frank W. R.
• Language: en_US
• Published: 2012
• Subjects: Bioactive compounds > Classification; Chromatophores; Biosensors; Computer network architectures; Cellular signal transduction
• Online Access: http://hdl.handle.net/1957/28942

Using network architecture to describe a biological system is an effective organizational method. The utility of this approach, which generally applies to qualitative models, is enhanced by the addition of quantitative models characterizing the interactions between network nodes. A chromatophore-based signal transduction network is developed, and the highly interconnected major nodes of the network, guanine trisphosphate, adenylate cyclase, and protein kinase A, are identified. These reference nodes serve to partition the network into functional modules, and mechanistic models describing these modules are derived. Three elicitor compounds, forskolin, melanocyte stimulating hormone (MSH), and clonidine, were selected due to their ability to access the signal transduction network at specific reference nodes, and the module configurations corresponding to their mechanisms of action are presented. The chromatophore responses to the three elicitors and to a negative control, L-15 cell medium, were recorded for two experimental blocks consisting of genetically different fish cells. Significant differences in cell responsiveness were evident between the two blocks, but this variability was controlled by the transformation and normalization of the data. The model parameters for each agent were estimated, and the resulting response curves were highly accurate predictors of the changes in apparent cell area, with R-squared values in the 0.88 to 0.96 range. Two examples were presented for the application of a model discovery algorithm, which selects modules from an existing library, generates model output for all valid module configurations, and selects the configurations which best satisfy a fitness function for a given set of target data. The algorithm proved robust to the introduction of different levels of random error in the simulated data sets when applied to a model of the desensitization of a cell membrane receptor, and continued to classify the stochastic data sets correctly even when the underlying rate constants differed significantly from those embedded in the modules. When challenged with the chromatophore data, the model discovery algorithm successfully matched the forskolin and MSH module configurations to the data within the top three models proposed, with less precise classification for the clonidine model. === Graduation date: 2005