Case-based Reasoning in Molecular Scene Analysis

In recent years, crystallographic studies have generated an explosion of information on molecular structure. The three-dimensional structural information obtained from such studies provides precise and detailed depictions of molecular scenes, an essential starting point for unraveling the complex rules of structural organization and molecular interactions in biological systems. This information is normally stored in databases in the form of unstructured atomic coordinate data. The research described in this paper is concerned with the organization of this structural knowledge into a case base of molecular scenes. The creation of such a case base provides a foundation for a computational strategy for the analysis of new scenes based on the experience embodied in the stored cases. Molecular scene analysis [FCG+93; GFA92] is concerned with the automated reconstruction and interpretation of crystal and molecular structures. A key feature of this research is the integration of the available knowledge on molecular structures into image reconstruction tools through the techniques of case-based reasoning and machine discovery. Discovered spatial and visual concepts index cases that can be used to anticipate the identity and relative locations of parts in a molecular scene. This paper describes how existing structural data can be organized so as to permit efficient and rapid retrieval from a case base of molecular scenes. Section 2 overviews the process of molecular scene analysis. In Section 3, issues involved in the representation, indexing, matching, adaptation and evaluation of molecular scenes are presented. The paper concludes with a discussion of related research in this area.