A Neural Network-Based QSAR Approach for Exploration of Diverse Multi-Tyrosine Kinase Inhibitors and its Comparison with a Fragment- Based Approach

Receptor and non-receptor tyrosine kinases have emerged as clinically useful drug target for treating certain types of cancer. It is well known that tyrosine kinase inhibitors with multi-kinases inhibitory potency are useful in anticancer therapy. In recent study, we have demonstrated application of a novel Group based QSAR (GQSAR) method to assist in lead optimization of multi-tyrosine kinase (PDGFR-beta, FGFR-1 and SRC) inhibitors. Although GQSAR method provides an alternative way to design new compounds, it could not be applied for virtual screening of large databases, because of its limitation to fragment each of the compound in the diverse database. So to circumvent this limitation of GQSAR method, herein we present the development of multi-kinase QSAR model using artificial neural networks. Various simple, easy and fast to calculate 2D/3D descriptors were used in the present analysis. The resulting neural network based QSAR (NN-QSAR) model was found to be statistically significant and provided insight into common structural requirements to inhibit different tyrosine kinases. The NN-QSAR model suggests five descriptors viz. number of rotatable bonds, number of hydrogen bond donors, number of building blocks, polar surface area and sum of nitrogen and oxygen atoms to be of major importance in explaining the activity variation in all the three kinases. In addition, this multi-target QSAR model could be useful to predict the activities of new compounds designed as tyrosine kinase inhibitors.