Abstract:
Insulin like growth factor (IGF) signaling system plays a critical role in regulating many
physiological processes such as cellular growth, transformation and apoptosis. IGF-1 receptor
(IGF-1R) plays a crucial role in cancer progression as its over expression induces the mitogenic
effects on binding to growth factors i.e. IGF-1 and IGF-2. Inhibition of tyrosine kinase domain of
IGF-1R in its basal state in various cancers has been associated with cell apoptosis. Various
strategies have been proposed to disrupt the IGF signaling pathway, but small molecule
inhibitors of tyrosine kinase domain have gained considerable attention due to higher
bioavailability, ease of synthesis and minimal side effects. Up to date, most of the inhibitors
against tyrosine kinase domain of IGF-1R have been identified from SAR studies but none of
them could reach the market due to toxicity, lack of efficacy and poor pharmacokinetics.
Therefore, it is very important to identify new arsenal of IGF-1R inhibitors with better
efficacy and reduced toxicity. Towards this goal, various molecular docking guided GRIND
models were developed, to explore the binding hypothesis and molecular basis of interaction of
tyrosine kinase domain of IGF-1R. Moreover, decision trees and artificial neural network models
were developed for the predictions and classification of already known inhibitors of tyrosine
kinase domain of IGF-1R into actives and inactives based on physicochemical attributes.
Molecular docking studies revealed that IGF-1R tyrosine kinase domain have the
potential to accommodate the structurally diverse compounds. Additionally, common scaffold
cluster analysis based on RMSD of common scaffold revealed three different binding positions
with in the active site. Briefly, the interaction with Glu 1080 of hinge region was observed
common in three identified clusters. Further, machine learning models were developed using
training set and cross validated by test set. Overall, neural network algorithm showed better
predictive accuracy rate of 96% compared to decision trees which showed 92% accuracy rate.
These models have the potential to make future predictions and virtual screening of new
chemical entities on basis of physicochemical descriptors against IGF-1R.
