Neural network based drug design for diabetes mellitus using QSAR with 2D and 3D descriptors

In this paper, we propose an artificial neural network approach to determine the quantitative structure-activity relationship (QSAR) among known aldose reductase inhibitors (ARI). In order to accurately describe the structural properties of ARIs, besides the popularly used 2-dimensional (2D) descriptors, we have used 3-dimensional (3D) molecular descriptors which are obtained through the DRAGON software. Multi-layer perceptrons (MLPs) with LM learning algorithm are used to determine the QSAR of the ARIs to predict two bioactivities, i.e., IC₅₀ and PI. We have shown that the performance of the proposed MLP-based model is much better in terms of RMSE and R-value than previous studies, which have used only two molecular descriptors, molar volume and electronegativity.