The expert system approach to predicting protein structure

Prediction of protein structure is an open-ended problem. Since an approach from first principles cannot be taken in reasonable computer time, short-cuts using further data are necessary. Such data include information about the specific protein in question, and information in databases which are about proteins in general. Is it possible to write a general, flexible 'superalgorithm' which would suit most circumstances? If so, it would seem likely to overcome one of the most understated but nonetheless greatest difficulties associated with molecular modelling and computer-aided drug design--reproducibility. To this end, a 'polymorphic programming environment' has been developed which represents both an expert system and a high-level language for theoretical chemists and molecular biologists. This language is GLOBAL (Ball et al., 1990). In a series of earlier studies, and more recently by means of GLOBAL itself, the nature of reproducibility and its rather surprising limits have been explored, and in general the current status and future potential of protein modelling have been examined.