Variable Selection by Recurent Neural Networks. Application in Structure Activity Relationship Study of Cephalosporins

摘要

Two methods for variable selection which are efficiently implemented by Hopfield -like Neural networks are described. Qualitative SAR models using the selected variables by both neural networks variable selection methods are built. The biological activity against Staphylococcus aureus of cephalosporins was used as dependent variable. The final correlation between observed and predicted activity values are good, indicating that the informative weight of the favored variables is high, providing a sound basis to select a good variable set of in qualitative structure-activity relationships (SAR) modeling.