Discovering Regulatory Network Topologies Using Ensemble Methods on GPGPUs With Special Reference to the Biological Clock of Neurospora crassa

Most genetic networks, such as that for the biological clock, are part of much larger modules controlling fundamental processes in the cell, such as metabolism, development, and response to environmental signals. For example, the biological clock is part of a much larger network controlling the circadian rhythms of about 2418 distinct genes in the genome (with 11 000 genes) of the model system, Neurospora crassa. Predicting and understanding the dynamics of all of these genes and their products in a genetic network describing how the clock functions is a challenge and beyond the current capability of the fastest serial computers. We have implemented a novel variable-topology supernet ensemble method using Markov chain Monte Carlo simulations to fit and discover a regulatory network of unknown topology composed of 2418 genes describing the entire clock circadian network, a network that is found in organisms ranging from bacteria to humans, by harnessing the power of the general-purpose graphics processing unit and exploiting the hierarchical structure of that genetic network. The result is the construction of a genetic network that explains mechanistically how the biological clock functions in the filamentous fungus N. crassa and is validated against over 31 000 data points from microarray experiments. Two transcription factors are identified targeting ribosome biogenesis in the clock network.