A Bchc Genetic Algorithm Model Of Cotemporal Hierarchical Arabidopsis Thaliana Gene Interactions

Gene interaction network models from time course gene transcript abundance data are algorithmically created using a new aggressive genetic algorithm denoted by BCHC. The BCHC algorithm rigorously integrates probabilistic hierarchical likelihood and Bayesian methodology to produce accurate posterior probabilities of interactions between genes after observance of hierarchical gene transcript abundance data. Forbidden pairwise gene relationships are incorporated into the modeling process. This gene interaction model is compared to a previous gene interaction model utilizing the same data and Bayesian likelihood, however based upon an exponentially slower, less aggressive, and less adaptive Metropolis-Hasting search algorithm. In addition for a smaller data set, our gene interaction model is compared to less rigorous non-probabilistic Lasso estimated partial correlation models which do not fully incorporate the hierarchical structure. A comparison is also made between the smallest Bayesian model and tests for edges based on a restricted non-Bayesian hierarchical technique. The BCHC algorithm performs well when the number of genes is moderately increased, both in terms of execution time and model quality.