Repairing inconsistent answer set programs using rules of thumb: A gene regulatory networks case study

Answer set programming is a form of declarative programming that can be used to elegantly model various systems. When the available knowledge about these systems is imperfect, however, the resulting programs can be inconsistent. In such cases, it is of interest to find plausible repairs, i.e. plausible modifications to the original program that ensure the existence of at least one answer set. Although several approaches to this end have already been proposed, most of them merely find a repair which is in some sense minimal. In many applications, however, expert knowledge is available which could allow us to identify better repairs. In particular, we consider the scenario where this expert knowledge is formulated as rules of thumb, but no training data is available to learn how these rules of thumb interact. The main question we address in this paper is whether we can then still aggregate the rules of thumb in a useful way. In addition to standard aggregation techniques, we present a novel statistical approach that assigns weights to these rules of thumb, by sampling, in a particular way, from a pool of possible repairs. In particular, we evaluate how frequently each given rule of thumb is violated in the sample of repairs, and use the Z-score of this distribution to set the weight of that rule. We analyze the potential of using expert knowledge in this way, by focusing on a specific case study: Gene Regulatory Networks. We describe the rules of thumb that express available expert knowledge from the biological literature and explain how they can be encoded while repairing inconsistencies. Finally, we experimentally compare the proposed repair strategies using rules of thumb against the baseline strategy of identifying minimal repairs.