Reasoning with imprecise probabilistic knowledge on enzymes for rapid screening of potential substrates or inhibitor structures

In many applications, there is a need to model and reason with imprecise probabilistic knowledge. In this paper, we discuss how to model imprecise probabilistic knowledge obtained from experiments in biological sciences on enzymes for rapid screening of potential substrate or inhibitor structures. Each imprecise probabilistic knowledge base is modelled as a probabilistic logic program (PLP). To predict a meaningful substrate structure, we have developed a framework (and a tool) in which a user (bioscientist) can query against a PLP (or a collection of PLPs), can examine how relevant a PLP is for answering a query, and can select a query result that is more satisfactory. This framework is implemented by integrating an optimizer in MatLab to solve the optimization problems subject to linear constraints. A preliminary version of the tool was demonstrated in the ECAI08 Demo session. Experimental results on evaluating the tool with probabilistic knowledge on enzymes for rapid screening of potential substrates or inhibitor structures demonstrate that this tool has a great potential to be used in many similar areas for the initial screening of compound structures in drug discovery.