Efficient Automatic Induction of Rules in Biological Systems
Mark E. Whiting, Philip R. LeDuc, Jonathan Cagan · The FASEB Journal · 2017
Rule based analysis of biological phenomena has rich potential but has been limited by our lack of methods for establishing systems of rules that discretely represent arbitrary structured data. Graph grammar is a formalism affording this kind of rule based representation of arbitrary information, and many tools have been developed to use graph grammars in engineering and design domains. However, inducting new graph grammar rules directly from a dataset is computationally complex and has not been automated; today it is done almost exclusively by hand, limiting the impact of the approach and mitigating its generalized use for analyzing biological phenomena. This work introduces a method for improving the complexity bound of this problem by using chunking and probabilistic matching of new rules, and a system architecture for facilitating the quick evaluation of rules in new and unknown datasets. The new method has been evaluated with engineering, architectural and statistical datasets and offers insight into rules about the construction, function and fundamental dynamics of systems. This work provides novel insights to areas from multiscale modeling to representations of biological phenomena.