The Evaluation of Multivariate Adaptive Regression Splines for the Prediction of Antitumor Activity of Acridinone Derivatives

Abstract

Multivariate adaptive regression splines (MARSplines) have been applied for the quantitative structure-activity relationships (QSAR) studies of antitumor activity of acridinone derivatives. Molecular modeling studies were performed with the use of HyperChem and Dragon software. The structures of the compounds were firstly pre-optimized with the MM+ mechanics and semi-empirical AM1 method procedure included in the HyperChem and resulting geometrical structures were studied with the use of Dragon software, and several molecular descriptors of acridinones were calculated and used as predictor (independent) variables in the MARS model building. Principal component analysis (PCA) was used to select the training and test sets. The optimal MARS model uses 28 basis functions to describe acridinones' antitumor activity and characterized high correlation between predicted antitumor activity and that one from biological experiments for the data used in the training and testing sets of acridinones with correlation coefficients on the level of 0.9477 and 0.9660, respectively. Generally, results showed that MARS model provided powerful capacity of prediction of antitumor activity of acridinone derivatives. Moreover, a physicochemical explanation of the descriptors selected by MARSplines analysis is also given, and indicated that molecular parameters describing 3-D properties as well as lipophilicity of acridinone derivative molecule are important for acridinones antitumor activity.