| Rapid technical advances in genomics and proteomics have catalyzed an extraordinary proliferation of experimental data. This offers an unprecedented opportunity to understand how organisms function, but also requires new analysis techniques. Existing methods encounter problems dealing with multiple data types, so we develop and present an event-based representation that overcomes these difficulties and also represents biological systems in a way that is compatible with notation used by experimental biologists. Given an ontology for a specific biological system, we define and characterize the space of possible regulatory hypotheses that can be made about that biological system. Taking inspiration from Stone Spaces, we show how the hypothesis space can (minimally) support the structure of a pre-topology. We present additional, biologically-motivated structure for the hypothesis space and show how to form classes of variant hypotheses that can be generated from a starting hypothesis by, for instance, "tweaking" its parameters, "borrowing" analogous portions from other biological systems, or "bridging gaps" in the explanation offered by the initial hypothesis. We introduce a control theory specifically designed for biological systems, and introduce and characterize the notion of "discoverability" of regulatory events. We characterize the decidability and discoverability properties of hypotheses and of variants generated by the perturbation operators, and show how to use them as part of a hypothesis "proofreading" system. We show how to model the hypothesis composition and proofreading process as a two-player game, and how algorithms can be developed to aid the human "player" in formulating and revising hypotheses. We develop a software prototype for such a hypothesis proofreading system and deploy the prototype for galactose metabolism in yeast. For this software, we use a novel contradiction-based approach that is more capable of handling large data sets than existing modeling approaches. To demonstrate this scalability, we expand our efforts to data from the human reactome. We show how to use the mathematical machinery we have developed to "proofread" the reactome, including testing it for consistency, completeness, precedence, enabling and supply. |
