This review provides a description of the background to the first international conference on Intelligent Systems for Molecular Biology and a problem-oriented overview of the papers presented. It focuses on genome analysis, gene identification, protein and RNA structure prediction and function, the modelling of biochemical pathways and data and knowledge bases. The range and quality of papers indicates that intelligent systems is emerging as an important sub-field of computational applications in the biological sciences.

Many areas of scientific endeavour can be characterized as the attempt to provide a consistent interpretation of a broad range of heterogeneous data and theories. In the area of protein structure prediction, for example, there are many types of diverse mutually constraining data and theories of protein structural organization that need to be integrated in order to produce a single consistent prediction (or set of predictions) of the protein structure from the experimentally derived amino acid sequence data. Understanding the role and function of proteins in the control of cell growth is an important part of contemporary cancer research. Protein structure prediction is immensely (combinatorially) complex, and traditional computational approaches to problems such as this have been based on the ‘generate and test’ paradigm in which hypotheses are first generated and then tested against any relevant constraints. In this paper we demonstrate the benefits of a new approach to solving large constrained combinatorial problems which uses the ‘constrain-and-generate’ paradigm and the ElipSys parallel constraint logic programming system. In ElipSys, constraints are used for a priori pruning of the search tree while parallelism enhances the efficiency of the remaining search. Initial results show several orders of magnitude increase in performance over a sequential logic programming (Prolog) approach. The improved performance can be attributed to the complementary actions of the constraint handling and support for parallelism in the ElipSys runtime system. Taken together, the additional performance and new knowledge representation techniques made possible using ElipSys significantly extend the range and complexity of scientific problems that can be addressed using logic programming languages.

An approach that accommodates classical decision theory within a framework of first-order logic with nonmonotonic extensions is described. Among the benefits offered by the approach are the potential: to express qualitative arguments about the desirability of decision options in the absence of probability or utility parameters; to automate techniques for generating decision options; and to initiate, control, and terminate a decision process autonomously. The rationality of a logical decision framework is discussed. A particular benefit of implementing decision procedures in a nonmonotonic logic is that decision systems can adapt automatically as beliefs change, contradictions are encountered, or new knowledge is acquired. A general conclusion is that metaknowledge and metalevel reasoning must be integral elements of a decision theory because of the versatility and operating flexibility they offer.>