Title: Robust Prediction with Explanatory Power for Protein Structure and Related Prediction Problems Objectives: This proposal aims to carry out an innovative and ambitious research programme to explore a series of novel and timely themes at the interface of various computer science traditions, with an impact on bioinformatics. Recent advances have demonstrated the potential for success in developing ever more general, intelligent and reliable classification and predictive technologies. We aim to explore and investigate the feasibility of designing and implementing robust classification and predictive systems that can deliver high quality predictions at the same time as human-understandable explanations for those predictions. The problem domain will be bioinformatics. A successful outcome to this challenging proposal could lead to a greater understanding of the issues that underpin predictive and classification software systems in general, and perhaps more importantly, those that have an immediate impact on quality of human life: bioinformatics. It is in domains like this where the availability of human-understandable explanations are as important as correct predictions. We propose to decompose the Protein Structure Prediction into a number of simpler problems and to tackle each one of these with a robust machine learning technique known as a Learning Classifier System (LCS). State-of-the-art LCSs can yield robust, accurate predictions, with the explanatory power not available in other methods. The methodology used to design representation and operators for the LCS will draw on past experience with cascade correlation neural networks (CCN). The contention of this proposal is that breaking the Protein Structure Prediction into these smaller problems, and providing a robust means of finding predictions with explanatory power, will yield information that will be difficult (if not impossible) to obtain with other methods and, ultimately, could lead to better holistic, i.e. re-constructing, approaches for structure prediction. Funding Body:EPSRC. Grant Reference: GR/T07534/01 Comments to jqb@cs.nott.ac.uk |