Machine Learning and Molecular Dynamics Simulations Predict Potential TGR5 Agonists for Type 2 Diabetes Treatment

*Table of Contents

• Background
• Objectives
• Methods
• Results
• Discussion
• Conclusion
• Team

*Background

Diabetes is a major health burden worldwide, and the development of novel and effective therapies is crucial for improving patient outcomes. A promising new class of medications for the management of type 2 diabetes is Takeda G Protein-Coupled Receptor 5 (TGR5) agonists. Since it has been demonstrated that activating TGR5 receptors improves glucose homeostasis and increases energy expenditure, these receptors are an attractive target for diabetes treatment. The development of novel TGR5 agonists holds great promise for diabetes treatment. These agonists have the potential to not only regulate glucose levels but also address other important aspects of diabetes management, such as body weight and associated complications. In addition, the stimulation of glucagon-like peptide 1 secretion by TGR5 agonists can lead to improved glycemic control through glucose-dependent insulin secretion.

Machine learning has emerged as a powerful tool in medical research, revolutionizing the field by enabling the prediction and diagnosis of various diseases, including diabetes. This has led to significant advancements in the monitoring and treatment of diabetes, allowing for personalized care and improved patient outcomes. By utilizing machine learning techniques, researchers have been able to analyze vast amounts of data related to diabetes, such as genetic information, environmental factors, and health behaviors. These techniques have led to the development of predictive models that can accurately identify individuals at risk for diabetes and provide personalized recommendations for treatment and management.

Machine learning-based models, molecular docking, and simulation techniques play a crucial role in the prediction of novel small molecules as potential TGR5 agonists in diabetes treatment. These advancements in machine learning have not only improved our understanding of diabetes but have also opened up new possibilities for the development of novel treatments. Machine learning algorithms have been applied to the discovery of potential therapeutic molecules for diabetes treatment. These algorithms have the capability to analyze molecular structures, predict their interactions with target proteins, and identify small molecules that can act as TGR5 agonists, which have shown promising potential for diabetes treatment. By combining machine learning models with molecular docking and simulation techniques we can efficiently screen large libraries of compounds and identify novel small molecules that have the potential to activate TGR5 and contribute to the treatment of diabetes.

*Objectives

1. To develop machine learning models to predict quantitative-structure activity relationship of the molecules to the target protein
2. Screen the COCONUT database to identify potential TGR5 agonist
3. Identify lead compounds by carrying out molecular docking to determine the binding affinity of the compound(s) to the target protein.
4. Determine the stability of the binding of the compound(s) to TGR5, through Molecular Dynamics Simulations

*Methods

Workflow

Figure 1. Workflow of the methods employed in this study

1.Machine Learning

The Jupyter Notebook was used to create and run the machine learning module's code. Bioactivity data was retrieved from the ChEMBL database (https://www.ebi.ac.uk/chembl/) The data was cleaned and classified as active, intermediate, or inactive.

The EC50 values were converted into pEC50 measurements, and Lipinski's descriptors were calculated. After the intermediate bioactivity category was removed, only active and inactive compounds were the subject of exploratory data analysis. Data matrices were created in accordance with the fingerprint descriptors produced using PaDEL. The dataset was split in an 80:20 ratio for training and testing after features with low variance were eliminated. Using a random forest algorithm, a regression model was produced, signifying the end of the machine learning training procedure.

The coconut database (https://coconut.naturalproducts.net/) was screened using the trained model, and the predicted dataset was used to prepare ligands for docking experiments against the TGR5 protein.

2.Molecular Docking

Molecular docking predicts the preferred orientation of a small molecule to a protein target to form a stable complex. Here, the TGR5 protein was downloaded from the protein data bank (https://www.rcsb.org/), and prepared using the protein preparation wizard of the Schrodinger software. A grid brox where the ligands were docked was created around the active site of the protein. The compounds predicted from the coconut database to have good binding with TGR5 were prepared using the LigPrep module of Schrodinger software. They were then docked into the active site of the previously prpared TGR5 protein.

3.Molecular Dynamics Simulation (MDS)

This is a computational method used to examine how atoms and molecules behave over time. Using Newton's equations of motion, MDS determines the stability, typically tracking the positions and velocities of atoms as they evolve in accordance with classical mechanics. This makes it possible for scientists to look into the atomic-level dynamic features of molecular systems. During system setup, the TGR5-ligand complexes from the docking studies were embedded in a POPC (300k) membrane bilayers. A suitable amount of ions was introduced to neutralise the system. After equilibration to stabilise the system's pressure and temperature, energy minimisation was done to eliminate any steric clashes. The protein-ligand complexes were solvated in an orthorhombic simulation box filled with explicit TIP3P water molecules. In order to prevent boundary effects, the buffer distance between the complex and the simulation box's edge was set to 10 Å. Additionally, 0.15 M NaCl was added to neutralise the system and replicate physiological conditions. The Desmond module of the Schrodinger software was used to run the simulation for a 100 nanosecounds duration; the simulation was run at a constant temperature of 300 K, pressure of 1 atm, and particle count. Prior to simulation, the model system was relaxed and allowed to equilibrate. Following this, a production run lasting 100 ns was executed, during which coordinates were recorded every 100 ps for further analysis. Throughout the run, the simulation trajectory was watched to guarantee system stability. MDS was run on a Linux operating system with GPU support. The TGR5-ligand complexes' conformational dynamics, binding stability, and interaction energy were evaluated by analysing the trajectory.

Using Schrödinger's Simulation Interaction Diagram tool, the simulation trajectories were examined in order to evaluate the stability and binding interactions between the ligand and the receptor. Important metrics analysed include:

• Root Mean Square Deviation (RMSD): To assess the protein-ligand complex's stability.
• Root Mean Square Fluctuation (RMSF): To evaluate each receptor residue's degree of flexibility.
• Radius of gyration (RoG): This is a measure of a ligand's extendendness that is comparable to its principal moment of inertia.
• The quantity of internal hydrogen bonds inside a ligand molecule is known as intramolecular hydrogen bonds, or intraHB.
• Ligand-Protein interactions: During the simulation, keep an eye on the different kinds of interactions (such as hydrogen bonds and hydrophobic contacts) that occur between the ligand and receptor.

*Results

Chemical Space Analysis

Fig 2: a. Frequency plot of the two bioactivity classes b. pEC50 values of active and inactive c. scatter plot of MW versus LogP

Fig 3: Box plots of active and inactive compounds analysed using Lipinski's descriptors

Fig 4: Scatter Plots of (a) Regression Model using Random Forest Algorithm (b) Experimental vs Predicted pEC50 for Training Data

Fig 5:a. Validation of the docking procedure. b.Predicted binding scores of compounds docked in the TGR5 binding pocket:CNP0209363, CNP0424850, CNP0417335, CNP0224616, and co-lig (INT-777) are given as -15.39, -14.87, -14.17, -14.01 and -9.008 kcal/mol, respectively

Fig 6:2D structures of lead compounds predicted from the COCONUT database (a. CNP0209363, b. CNP0424850, c. CNP0417335, d. CNP0224616)

Fig 7:Molecular binding interactions of compounds docked in the TGR5 binding pocket

Fig 8: Trajectories of the Root Mean Square Deviation (RMSD) of TGR5 in apo state, in complex with the Co-lig (INT-777) and the lead compounds

Fig 9: Trajectories of the Root Mean Square Fluctuation (RMSF) of TGR5 in apo state, in complex with the Co-lig (INT-777) and the lead compounds

Fig 10: Radius of gyration (RoG) of TGR5 in complex with the compounds

Fig 11: Intrahydrogen bonds present in the compounds

Fig 12: Protein-Ligand contact during MD simulation between TGR5, Co-lig and the lead compounds

Fig 13: Ligand-Protein contact made during MD simulation between TGR5, Co-lig and the lead compounds a.Co-Lig; b. CNP0209363; c. CNP0424850; d. CNP0417335; e. CNP0224616

Discussion

**The bioactivity classes are spanning similar chemical spaces as evident by the scatter plot of MW vs LogP. **Only LogP showed no difference between the actives and inactives among the four Lipinski's descriptors (MW, LogP, NumHDonors, and NumHAcceptors); the other three descriptors (MW, NumHDonors, and NumHAcceptors) all show statistically significant differences between the actives and inactives. **The lead compounds show better binding to TGR5 compared to the co-crystallized ligand as displayed by the docking scores **Overall, the lead compounds were quite stable within the binding pocket of TGR5 as revelaed by the low RMSD values they displayed.

*Conclusion

Our study identified four (4) novel molecules, notably, CNP0209363, CNP0424850, CNP0417335 and CNP0224616 from the COCONUT database, as potential TGR5 agonists The findings of this study provide valuable supporting data for predicting potential TGR5 agonists that combines machine learning-based models, molecular docking, and molecular dynamics simulations, which demonstrate the potential of the proposed approach for predicting novel small molecules as TGR5 agonists that could be considered for the treatment of T2D .

*Team

• Ojochenemi Enejoh (Genomics and Bioinformatics Department, National Biotechnology Development Agency, Abuja, Nigeria) [email protected]
• Hector Nortey (Department of Clinical Pathology, Noguchi Memorial Institute for Medical Research, College of Health Science, University of Ghana, Legon, Accra, Ghana) [email protected]
• Chinelo Okonkwo (Department of Pharmacy, National Hospital Abuja, Nigeria) [email protected]
• Ainembabazi Moses (African Centers of Excellence in Bioinformatics, Department of Immunology and Microbiology, Makerere University, Kampala, Uganda) [email protected]
• Olalekan Kemiki (Molecular and Tissue Culture Laboratory, Babcock University, Ilisan-remo, Ogun State, Nigeria) [email protected]
• Florence Mbaoji (Department of Pharmacology and Toxicology, Faculty of Pharmaceutical Sciences, University of Nigeria, Nsukka, Enugu State, Nigeria.) [email protected]
• Abdulrazak Sale (Department of Biochemistry, Faculty of Basic Health Science, Bayero University, Kano, Nigeria) [email protected]
• Olaitan I Awe (African Society for Bioinformatics and Computational Biology, Cape Town, South Africa) [email protected]