Switching Between Discrete and Continuous Models To Predict Genetic Activity

Abstract

Molecular biologists use a variety of models when they predict the behavior of genetic systems. A discrete model of the behavior of individual macromolecular elements forms the foundation for their theory of each system. Yet a continuous model of the aggregate properties of the system is necessary for many predictive tasks. I propose to build a computer program, called PEPTIDE, which can predict the behavior of moderately complex genetics systems by performing qualitative simulation on the discrete model, generating a continuous model from the discrete model through aggregation, and applying limit analysis to the continuous model. PEPTIDE's initial knowledge of a specific system will be represented with a discrete model which distinguishes between macromolecule structure and function and which uses five atomic processes as its functional primitives. Qualitative Process (QP) theory [Forbus 83] provides the representation for the continuous model. Whenever a system has multiple models of a domain, the decision of which model to use in a given time becomes a critically important issue. Knowledge of the relative significance of differing element concentrations and the behavior of process structure cycles will allow PEPTIDE to determine when to switch reasoning modes.