The quotient space theory of problem solving
Fundamenta Informaticae - Special issue on the 9th international conference on rough sets, fuzzy sets, data mining and granular computing (RSFDGrC 2003)
Prediction of Protein-Protein Interaction Sites Using Granularity Computing of Quotient Space Theory
CSSE '08 Proceedings of the 2008 International Conference on Computer Science and Software Engineering - Volume 01
Hi-index | 0.00 |
Predicting of protein-protein interaction sites (PPIs) merely are researched in a single granular space in which the correlations among different levels are neglected. In this paper, PPIs models are constructed in different granular spaces based on Quotient Space Theory. We mainly use HSSP profile and PSI-Blast profile as two features for granular space, then we use granularity synthesis theory to synthesis PPIs models from different features, finally we also improve the prediction by the number of neighboring residue. With the above method, an accuracy of 59.99% with sensitivity (68.87%), CC (0.2113), F-measure (53.12%) and specificity (47.56%) is achieved after considering different level results. We then develop a post-processing scheme to improve the prediction using the relative location of the predicted residues. Best success is then achieved with sensitivity, specificity, CC, accuracy and F-measure pegged at 74.96%, 47.87%, 0.2458, 59.63% and 54.66%, respectively. Experimental results presented here demonstrate that multi-granular method can be applied to automated identification of protein interface residues.