Bayesian method for biological pathway discovery from high-throughput experimental data

This paper describes a novel Bayesian method for discovering intra-cellular pathways from high throughput data. This Bayesian method is generalized from a deterministic algorithm, and it combines experimental data with prior belief to produce as output a probability distribution over the possible causal relationships between each pair of variables. We applied this algorithm to gene expression data on galactose metabolism in yeast. The area under ROC curve (AUROC) for the Wagner algorithm is 0.64. For the Bayesian algorithm the AUROC is 0.87, with a 95% confidence interval of (0.77 0.94). Thus, the Bayesian algorithm performs statistically significantly better than Wagner algorithm.