Learning Causal Networks from Gene Expression Data

In this work we present a new model for identifying dependencies within a gene regulatory cycle. The model incorporates both probabilistic and temporal aspects, but is kept deliberately simple to make it amenable for learning from the gene expression data of microarray experiments. A key simplifying feature in our model is the use of a compression function for collapsing multiple causes of gene expression into a single cause. This allows us to introduce a learning algorithm which avoids the over-fitting tendencies of models with many parameters. We have validated the learning algorithm on simulated data, and carried out experiments on real microarray data. In doing so, we have discovered novel, yet plausible, biological relationships.