Computational Investigation of Glucose Binding Receptor

Abstract

The rapid rise in diabetes patients worldwide demands new diabetes diagnostics and therapeutics. Targeting glucose in the human body could be the future insulin alternative therapy. However, glucose recognition in an aqueous solution is a challenging task. The hydroxyl groups present on glucose molecules resemble hydroxyl groups present on water molecules and hide in the solvent, making it challenging to distinguish hydroxyl groups of glucose from water. Even the natural carbohydrate-binding proteins such as Lectins have a low affinity towards glucose. All though glucose recognition in water is not impossible. Several proteins and boronic acid-based receptors are developed for glucose sensing. The structural instability of proteins in abnormal environmental conditions and low affinity of boronic acid receptors forced the search for glucose selective synthetic receptors. Professor Anthony Davis and his team from the University of Bristol developed a glucose selective synthetic receptor by using three [2-(Carbamoylamino)phenyl]urea pillars as polar fragments and a pair of triethyl-mesitylene as an apolar fragment. The receptor is developed based on temple architecture that can encapsulate the glucose molecule in its cage binding through a series of hydrogen bonding and CH-π interactions with high affinity in a solvent using the rational method. However, rational molecular designing is slow and limited to human ideas. Therefore, high-throughput virtual screening has been carried out in this work to identify the glucose binding fragments from the ZINC compound database using the GLIDE program. Ideal fragments are selected based on the glide score. Further, the binding affinity of the glucose-compound complexes was calculated using the MM-GBSA method. Nevertheless, GLIDE and MM-GBSA are compute-intensive physics-based computational methods that limit the exploration of diverse compound databases. Hence, there is a need for high-speed techniques to screen compounds from databases. The machine learning algorithms have been implemented to develop classification and regression prediction models using high-throughput virtual screening and MM-GBSA results. Further, an algorithm was developed inspired by the solar system and named the “astrophysics-based” algorithm. Machine learning algorithms include Decision Tree, Support Vector Machines, Random Forest, K-Nearest Neighbour, Lasso Regression, Elastic Net, Ridge Regression, Stochastic Gradient Descent, and astrophysics-based algorithm, which are implemented to develop the glucose binder and binders’ affinity prediction models. The developed prediction model helps screen the compounds from various databases within less time with reasonable accuracy. Furthermore, the top-ranked fragments from the high throughput screening and machine learning workflow are taken to construct symmetric synthetic glucose receptors. The receptors are modeled based on three-pillar temple architecture with a pair of triethyl-mesitylene(apolar/nonpolar) fragments regarded as the receptor’s ceiling and floor. The roof and floor of the receptor are connected with the polar fragments from virtual screening and the ML model as pillars. From screened compounds, it is observed that MR3 designed receptor resembles the synthetic receptor from literature, proving that our study is significant for developing novel synthetic glucose receptors. The designed receptors are geometrically optimized with molecular mechanics. The glucose docked with the receptors and the complex structures are subjected to molecular dynamics simulations, and finally, the glucose selectivity of the designed receptor was predicted by comparing with galactose and mannose monosaccharides. Computer experimental results reveal that virtual screening is one of the methods which screened the compounds with more nitrogen atoms that form hydrogen bonds with glucose. On the other hand, the proposed astrophysics-based framework provides 99.30\% accuracy in classifying glucose binder, and predicting binding affinity with 0.04 root mean square error (RMSE). 5-fold cross-validation results showed that the proposed astrophysics-based model delivers better accuracy than other machine learning models. The projected research workflow enables researchers to screen glucose binders rapidly from the extensive compound library with less computational time for designing synthetic glucose receptors. The docking results reveal that the receptor models MR1, MR2, and MR3 based on ZINC82047919, ZINC238094340, and ZINC238519600 compounds show better interactions and glucose orientation within the receptors cavity. The molecular dynamics simulation results revealed that the MR1 and MR2 receptors showed better interaction and selectivity towards glucose over galactose and mannose. The MR2 receptor designed using compound ZINC238094340 could not hold the glucose and undergo significant conformation changes during the simulation process.