Name
Abstract | ||
|---|---|---|
Rough set algorithm was used as a new methodology to build structure-activity relationship (SAR) models in this paper. It acted as feature selector and nonlinear rule generator. The SAR model expressed as human readable if-then rules was developed for the inhibition of the serine/threonine kinase CDK1/cyclinB by compounds from the indirubin inhibitor family. The feature selection ability of rough set algorithm was compared with the build-in approaches (CfsSubsetEval and ConsistencySubsetEval) in Weka under leave-one-out (LOO) and 10-fold cross-validation. Through training a set of 31 objects, a rule-based SAR model had been built with a reduct generated by rough set. The predictability of the model was evaluated by an external test set of 16 compounds. The existing powerful approaches, such as the decision tree learners, neural network, support vector classifier and LogitBoost approaches, were used to verify the performance of rough set method. It revealed that rough set method should play important role in data preprocessing and model building of nonlinear SAR analysis. The advantages and limitations of rough set-based SAR analysis were discussed. The results were satisfactorily in accordance with the available understanding of cocrystal structures and 3D QSAR models. |
Year | DOI | Venue |
|---|---|---|
2010 | 10.1016/j.eswa.2009.12.008 | Expert Syst. Appl. |
Keywords | Field | DocType |
nonlinear sar analysis,feature selection ability,rough set,rough set-based sar analysis,model building,rule-based sar model,sar model,rough set method,inductive method,feature selection,rough set algorithm,cyclin-dependent kinase,structure–activity relationship,indirubins,qsar model,neural network,structure activity relationship,rule based,decision tree,data preprocessing,cross validation,cyclin dependent kinase | Data mining,Decision tree,Reduct,Feature selection,Computer science,Data pre-processing,LogitBoost,Artificial intelligence,Artificial neural network,Pattern recognition,Rough set,Machine learning,Test set | Journal |
Volume | Issue | ISSN |
37 | 7 | Expert Systems With Applications |
Citations | PageRank | References |
3 | 0.37 | 5 |
Authors | ||
5 | ||