Guiding revision of regulatory models with expression data.

BioLingua is a computational system designed to support biologists' efforts to construct models, make predictions, and interpret data. In this paper, we focus on the specific task of revising an initial model of gene regulation based on expression levels from gene microarrays. We describe BioLingua's formalism for representing process models, its method for predicting qualitative correlations from such models, and its use of data to constrain search through the space of revised models. We also report experimental results on revising a model of photosynthetic regulation in Cyanobacteria to better fit expression data for both wild and mutant strains, along with model mutilation studies designed to test our method's robustness. In closing, we discuss related work on representing, discovering, and revising biological models, after which we propose some directions for future research.