Some 270 delegates attended the Third International Conference on Intelligent Systems for Molecular Biology (ISMB-95) held in Cambridge, England, on July 16-19, 1995. The conference brought together scientists who are using advanced computational methods to address problems in molecular biology. These methods include data modeling, machine learning, artificial intelligence, cognitive science, robotics, combinatorial and stochastic optimization, adaptive computing, string and graph algorithms, linguistic methods, and parallel computer technologies.

The conference was preceded by 8 introductory and advanced tutorials attended by 187 delegates. The best-attended tutorials were (1) Protein Structure Prediction and (2) Statistical Foundations of Multiple Sequence Alignments and Structure Prediction.

The conference consisted of 8 sessions and 26 oral papers on key bioinformatics issues: protein structure and docking, sequence alignment, protein sequence and structure, understanding sequence function, RNA sequence and structure, genome information systems, gene finding and gene structure, and database searching. Topics of particular interest included methods for automatically identifying different structural and functional domains within protein structures and the use of advanced statistical methods (e.g., hidden Markov models and stochastic context-free grammars) for identifying complex protein sequence patterns and sequence relationships. Recurring themes were (1) genome-information integration and presentation and (2) methods for rapidly searching DNA and protein sequence and structure databases for related molecules.

Throughout the meeting, two general trends could be discerned: (1) the need to compare new approaches to existing methods, with the associated complexities of selecting appropriate data sets, and (2) the desire to make predictions of biological function that could be validated properly. This raises the general question of how to represent biological function in molecular biology databases. More complete databases of biological processes and pathways that can be linked to sequence and structure data sets are clearly needed.

A new topic was the search for efficient and accurate methods of finding the best fit between macromolecular structures (docking), with both protein-protein and protein-ligand interactions being considered. Several papers and a tutorial covered the challenges of protein structure prediction and the refinements to methods for threading sequences through known protein structures. Applications for machine-learning methods (e.g., neural networks, hidden Markov models, and genetic algorithms) were well represented, and a new approach based on a simulation of the immune system was presented. RNA structure prediction was addressed by several different techniques, including the use of linguistics and graph theory.

Conference proceedings (ISBN 0-929280-83-0), including 26 oral papers and 22 posters featured in the formal poster session, can be obtained directly from AAAI/MIT Press (info@aaai.org). Proceedings of ISMB-93 (ISBN 0-929280-47-4) and ISMB-94 (ISBN 0-929280-68-7) are also available from AAAI. ISMB-96, being organized by David States, will be held June 12-15 at Washington University, St. Louis.