Predicting the free radical scavenging activity of curcumin derivatives

A data set of 22 curcumin derivatives with DPPH free radical scavenging activity was used for classification and quantitative structure-activity relationship (CSAR and QSAR) study. Geometry optimization was performed at B3LYP/6-31g(d) level to generate descriptors based on electronic properties, which comprised of dipole moment, hardness, softness, energy difference of highest occupied molecular orbital energy (HOMO) and lowest unoccupied molecular orbital energy (LUMO). CSAR models were constructed using partial least squares (PLS) and support vector machine (SVM) methods for classifying compounds based on their antioxidant activity as a function of the calculated descriptors. Descriptors based on structural property (e.g. number of hydroxyl groups) and electronic properties were shown to be important in classifying the compounds. The PLS and SVM models were 100% accurate. The descriptors were further employed in the development of QSAR regression model using PLS, multiple linear regression (MLR), and SVM. Various data sampling approaches and statistical parameters were employed to assess the predictivity and validity of the developed models. CSAR models achieved accuracies in the range of 84.21 to 100% while QSAR models exhibited correlation coefficients in the range of 0.942 and 0.999 along with root mean square error between 0.108 and 0.175. In both CSAR and QSAR studies, SVM was the best performing model for predicting the antioxidant activity of curcumin derivatives. The models described herein have great potential for the rational design of novel curcumin derivatives with promising free radical scavenging activities. Particularly, it was observed for high activity compounds that the chemical stability was high as suggested by the lower hardness, higher softness and higher HOMO–LUMOgap values than those of low activity compounds. Moreover, high activity compounds also possessed lower dipole moment value and higher number of hydroxyl groups than that of low activity compounds.

► QSAR model for classifying and predicting antioxidant activity of curcumin analogs.
► Molecular descriptors accounted for substituent effects and electronic properties.
► Potent compounds have high chemical stability and OH count with low dipole moment.
► SVM was the best performing model; outperforming MLR and PLS.
► Several cross-validation and external validation confirm robustness of QSAR models.