Analogy-based hypothesis generation combined with ontology-based deduction is a promising technique for knowledge discovery and validation. We are using this combined approach to improve the quality of analogy reasoning. This paper is a report of our work in progress in that direction. We will discuss the formal basis and method of the approach from a symbolic machine-learning point of view and propose a generalized model for analogy-based hypothesis generation that allows multi-strategy learning of analogies. We will also present the results of our experiments using this combined approach with the unstructured summary data from the Center for Nonproliferation Studies (CNS) and discuss possible improvements. Finally, we will propose some research issues in order to further develop and deploy this technique.