VIETNAM NATIONAL UNIVERSITY, HANOI VIETNAM JAPAN UNIVERSITY ——————— NGUYEN VIET ANH INVESTIGATE OF INTERACTION OF MU-OPIOID RECEPTOR WITH UNBIASED AND BIASED LIGANDS USING MOLECULAR
Motivation
Every year, a large number of prescriptions for pain management medications, particularly opioids, are issued While opioids are effective analgesics used in medical treatments, they carry serious side effects such as constipation, nausea, vomiting, respiratory depression, and a risk of addiction The opioid crisis has led to numerous overdose fatalities annually, highlighting the urgent need for the development of new painkillers that offer effective relief with fewer adverse effects.
Opioids exert their analgesic effects by binding to opioid receptors, which are part of the G protein-coupled receptor superfamily The three primary opioid receptors—mu (μOR), delta (δOR), and kappa (κOR)—serve as targets for medications addressing mood and pain The mu-opioid receptors (MORs) are particularly significant in pain relief and are the main focus of painkiller research Studies on MOR mutant mice indicate that opioid binding to these receptors can produce various effects across different organ systems beyond just pain relief These effects may include adverse reactions associated with both short-term and long-term opioid use, such as respiratory depression, gastrointestinal motility reduction, nausea, vomiting, dizziness, and hallucinations, alongside a state of calmness and relaxation.
Long-term opioid use leads to persistent activation of mu-opioid receptors (MORs) and associated G-protein signaling pathways, which can cause significant changes in the body, including tolerance, hyperalgesia, and physical dependence While MORs enhance the pleasurable and euphoric effects of opioids, their prolonged activation contributes to these adverse effects.
The US Centers for Disease Control and Prevention has identified opioid misuse as a critical public health crisis over the past decade, leading to numerous poisoning incidents In 2015 alone, over 33,000 individuals in the United States died from opioid overdose, highlighting the severity of this epidemic.
[27] Opioid misuse resulted in an estimated 50,000 fatalities in 2017 Between
2001 and 2016, there were more than 335,000 fatalities caused by opioids [21] In
In 2021, approximately 296 million people, or 5.8% of the global population aged 15-64, engaged in drug use, with around 60 million specifically using opioids That year, about 39.5 million individuals suffered from drug use disorders, predominantly relying on illegally produced heroin, although the number of those addicted to prescription opioids is rising Recent advancements in safer opioid analgesics have emerged, with a notable strategy called "biased agonism" gaining attention.
Similar to other G protein-coupled receptors (GPCRs), the primary signaling pathways of the opioid receptor (àOR) involve G proteins and β-arrestins, each playing unique biochemical and physiological roles It is commonly understood that most known opioids activate both G protein and β-arrestin signaling pathways equally upon binding to àOR However, biased agonists can selectively target the signaling route that provides analgesic benefits while minimizing adverse effects such as constipation, respiratory depression, and addiction.
Hence, it is extremely important to comprehend the binding conformations of
From 2000 to 2014, the United States experienced notable mortality rates linked to prescription opioids and heroin drug poisoning This analysis aims to explore the differences in the structures of opioid receptors (àOR) and their ligands, both biased and unbiased, to understand how these variations lead to the activation of different signaling pathways The primary focus will be on investigating the molecular-level interactions of various poses that play a critical role in opioid binding.
Objectives
The primary objective of this thesis is to investigate, examine, and assess molec- ular docking using machine learning and deep learning models The goals may be succinctly described as:
1 Overview of protein-ligand binding, molecular docking techniques, virtual screening, and de novo drug design
2 Evaluate a deep learning approach for virtual screening that is capable of identifying compounds of the opioid family with a strong affinity for their receptors and verifying their effectiveness
3 Evaluate a deep learning approach for molecular docking that is capable of identifying poses with a strong affinity for their receptors and verifying its effectiveness.
Thesis outline
This article is organized into six sections: overview, methodologies, case studies, results, conclusion, and supplementary materials The project intersects biochemistry, biophysics, and machine learning, highlighting the importance of interdisciplinary collaboration To bridge the knowledge gap among specialists from diverse fields, a thorough explanation of the essential procedures and concepts is provided.
This article explores the modern drug research process, highlighting the significant role of artificial intelligence in drug design It outlines the key characteristics and foundational aspects of artificial intelligence, as well as the current advancements in the field.
This thesis explores the essential elements of molecular docking and molecular dynamics modeling, highlighting their significance as key research techniques It details various components and their characteristics, focusing on their benefits, drawbacks, and real-world applications Additionally, the article reviews recent advancements in docking-based molecular screening and discusses the practical implications of both techniques.
PROTEIN, LIGAND AND BINDING AFFINITY
Protein
Proteins are intricate biological molecules made up of long chains of amino acids, essential for various functions in living organisms They play crucial roles in metabolic processes, DNA replication, response to stimuli, and the formation of cellular structures, as well as transporting chemicals The diversity in protein functions arises from variations in their amino acid sequences, which are dictated by the genetic code This sequence ultimately influences how proteins fold into specific three-dimensional structures, determining their specific roles in biological systems.
A polypeptide is a linear sequence of amino acid residues, and a protein is defined as having at least one extended polypeptide Peptides, which are shorter chains of fewer than 20-30 residues, are typically not considered proteins The amino acid sequence of a protein is dictated by the gene sequence encoded in the genetic code, which includes 20 standard amino acids After synthesis, proteins undergo post-translational modifications that alter their physical and chemical properties, stability, activity, and overall function Some proteins also contain non-peptide components known as prosthetic groups or cofactors Additionally, proteins can work together to perform specific functions, often forming stable protein complexes.
Proteins are essential building blocks of living organisms, integral to nearly every cellular function Many proteins act as enzymes, enabling vital biological processes and significantly contributing to metabolism.
Figure 2.1: Mu Opioid Receptor-Gi Protein Complex [4]
Figure 2.2: Structure of an amino acids [2] have crucial roles in cell signaling, immunological responses, cell adhesion, and the cell cycle
Protein molecules can be classified into two main categories: fibrous proteins and globular proteins Fibrous proteins are elongated and insoluble, while globular proteins are compact and soluble.
Figure 2.3: Structure of 20 amino acids [2] structure Fibrous and globular proteins may exhibit one of four kinds of protein structures, namely primary, secondary, tertiary, and quaternary structures
The primary structure of a protein refers to the unique arrangement of amino acids in a polypeptide chain, characterized by the amino terminus (N-terminus) and carboxy terminus (C-terminus) based on their terminal free radicals The sequence of amino acid residues is always counted starting from the N-terminus This specific amino acid sequence, determined by genetic information, is crucial for understanding the protein's structure and function.
2 Secondary structure refers to the specific three-dimensional conformation of a small region of a protein Hydrogen connections between atoms along the polypeptide chain’s backbone create them
The tertiary structure of proteins is defined by the three-dimensional arrangement of all atoms within a single polypeptide chain, held together by stabilizing interactions between the side chains and backbone groups Key stabilizing interactions in this structure include disulfide bonds and salt bridges.
Figure 2.5: Secondary protein structure [3] coordination connections with metal ions, hydrogen bonds, and hydrophobic interactions
4 Quaternary structure refers to the organization of numerous folded protein
Figure 2.6: Tertiary protein structure [1] subunits inside a complex consisting of multiple subunits
Proteins are vital functional molecules in all living organisms and play a crucial role in therapeutic development against diseases Small organic compounds, derived from natural sources or synthesized in laboratories, can effectively inhibit biologically active enzymes and receptors involved in disease processes Understanding the 3D structure of a receptor allows researchers to utilize computational techniques, such as virtual screening and docking, to explore a wide range of chemicals from chemical databases for potential inhibitors The protein-ligand docking process involves several steps, including identifying active sites, assessing ligand flexibility, and evaluating interaction energy between the ligand and protein In cases where 3D structures of receptors are unavailable, homology modeling can be employed to create a 3D model for subsequent virtual screening and docking.
Ligand
A ligand, in the field of biochemistry, refers to a molecule that has the capability to attach to a biomolecule and create a complex in order to fulfill a biological func-
An effector molecule is a substance that binds to a specific site on a target protein through intermolecular interactions, including ionic bonds, hydrogen bonds, and Van der Waals forces This docking process, which refers to the association of the effector with the protein, is typically reversible, allowing for dissociation.
The receptor, which is a protein in this instance, is usually a large and rather immobile molecule that has a distinct binding site for the ligand
Ligands spread throughout the surrounding area until they bind to specific receptors, triggering a change in the receptor protein's three-dimensional structure This conformational change determines the receptor's functional state, while affinity describes the strength of the binding interaction Ligands encompass a diverse range of molecules, such as substrates, inhibitors, activators, and neurotransmitters.
Ligand-Receptor Interactions
In pharmacology, the interaction between ligands and receptors is crucial, as each receptor is activated by specific endogenous ligands like hormones Agonism occurs when a ligand binds to a receptor, triggering a biological response Pharmaceuticals often utilize endogenous agonists or their derivatives, such as growth hormone and adrenaline, to achieve therapeutic effects.
Partial agonists activate receptors but produce a weaker response compared to full agonists Despite their differences in efficacy, both full and partial agonists often share similar chemical structures.
Antagonists are molecules that attach to a receptor, but they do not trigger a reaction Antagonists effectively obstruct the receptor, preventing the agonists from binding and hence impeding the agonists’ activity.
Binding Affinity
Binding affinity refers to the intensity of the binding relationship between a certain biomolecule, such as a protein or DNA, and its ligand or binding partner, such as a medication or inhibitor
The law of mass action in chemistry asserts that the rate of a chemical reaction is directly proportional to the concentrations or activities of the reactants This principle clarifies and predicts the behavior of solutions in dynamic equilibrium, indicating that the ratio of reactant and product concentrations remains constant in a chemical reaction at equilibrium.
The interaction between ligands and receptors follows the rule of mass action,
The relationship between overall concentration and the rate of molecular processes is crucial, as it allows for the determination of equilibrium concentrations of bound receptors through the rates of synthesis and degradation.
The ligand-receptor interaction may occur in two states, as shown by the reac- tion scheme below: where:
- [L], [R] is the molar concetrations of free ligands, receptors at equilirium, unit is M
- [LR] is the molar concentration of complex, unit is M
The rate constants K on and K off indicate the formation and dissociation speeds of the ligand-receptor complex It is believed that after detachment, both the receptor and ligand remain unchanged.
Typically, the equation is based on the following assumptions:
• The interaction may be reversed
• The receptor, ligand, and ligand-receptor complex exist in a state of equilib- rium
• The receptor has a single binding site for the ligand
• The ligand and receptor swiftly interact to create the ligand-receptor complex
Temperature affects the rates of association and dissociation in chemical reactions, while the concentration of reactants also plays a significant role Equilibrium is reached when the formation and dissociation rates of ligand-receptor complexes are equal.
At equilibrium, the following conditions are true:
The dissociation constant K d is the inverse of the receptor’s affinity for the ligand at equilibrium
The dissociation constant of ligands typically ranges from mM to nM, and the binding affinity, represented as pK = −log(Kd), is commonly used to quantify the strength of this binding.
The affinity between a ligand and its receptor indicates the strength of their binding interaction An increase in the Kd value or a decrease in the pK value signifies a reduction in binding strength and affinity Conversely, a ligand exhibiting a low Kd or a high pK demonstrates stronger binding and higher affinity.
Modern drug design
Drug discovery involves the systematic analysis of compounds to identify their medicinal properties and potential uses This process is both time-consuming and costly, with average expenses reaching 1.3 billion USD and a timeline of 10 to 15 years for completing all clinical trial phases Notably, 90% of drug discovery initiatives fail, and the costs associated with drug development continue to increase over time.
Reevaluating traditional drug design methods can provide an alternative approach to counteract current challenges in the field Historically, drug discovery has depended on extensive laboratory testing of chemical compounds, which remains costly despite technological advancements However, recent breakthroughs in computer-aided drug design have expanded options significantly, allowing for the rapid screening of larger compound libraries through digital methods The integration of artificial intelligence has further facilitated the creation of new molecules, while time-consuming and expensive laboratory tests can often be substituted with faster, more cost-effective in silico techniques.
Bioinformatics plays a crucial role in pharmaceutical research, extending beyond small compounds to enhance drug delivery systems, biopharmaceuticals, and antibodies by improving their specificity, affinity, and stability Structural analysis offers insights into the functions and characteristics of proteins and small molecules, which can be modified with specialized software Tailored simulations are employed to evaluate interactions and overall performance, making bioinformatics an indispensable tool in drug development.
Recent advancements in pharmaceutics have been driven by the integration of artificial intelligence, particularly through deep learning models A notable example is AlphaFold, which can accurately predict the three-dimensional structures of proteins based solely on their amino acid sequences Major pharmaceutical companies such as Roche, Johnson & Johnson, and Pfizer are increasingly interested in leveraging AI for drug discovery A successful case of this technology in action is Paxlovid, Pfizer’s oral medication for COVID-19, which was developed using machine learning techniques.
Protein case studies
G Protein-coupled receptors
G-protein-coupled receptors, also known as GPCRs, are the most extensive and varied collection of membrane receptors found in eukaryotes Cell surface recep- tors transmit signals they receive via guanine nucleotide-binding proteins, often known as G-proteins These cell surface receptors function as receptors for sev- eral types of stimuli, such as light energy, peptides, lipids, carbohydrates, and proteins These signals provide cells with information about the availability or lack of essential light or nutrients in their surroundings, or they transmit informa- tion from other cells Humans possess around 1,000 distinct GPCRs, with each receptor exhibiting a high degree of specificity towards a single signal Despite the vast number of GPCRs, all have a common fundamental structure: a solitary polypeptide chain that weaves across the lipid bilayer of the plasma membrane in a pattern of seven loops Due to this rationale, they are sometimes referred to as seven-pass transmembrane (7TM) receptors
G protein-coupled receptors (GPCRs) play crucial roles in various physiological processes within the human body, and recent advancements in our understanding of these receptors have greatly influenced modern medicine It is estimated that approximately 50% of all pharmaceuticals available today exert their effects by binding to GPCRs.
Figure 2.8: The polypeptide has one end that forms the extracellular domain responsible for binding the signal, while the other end is located in the cytoplasm of the cell [16]
When a ligand binds to the extracellular domain of a GPCR, it triggers a structural change in the receptor, allowing it to interact with a G-protein, which subsequently relays the signal to other molecules in the signaling pathway.
à-Opioid Receptors
Opioid receptors (ORs), a type of G protein-coupled receptors (GPCRs), play a crucial role in the body's response to hormones, neurotransmitters, and medications, influencing sensory experiences such as vision, taste, and smell They are primarily responsible for the analgesic effects of opioids, which interact with these receptors to produce both inhibitory and excitatory outcomes at neural synapses Key medications that target mood and pain often focus on three specific opioid receptors: the μ, δ, and κ-opioid receptors.
1 à receptors are mostly located in the brainstem and medial thalamus à receptors mediate supraspinal analgesia, respiratory depression, euphoria, se- dation, reduced gastrointestinal motility, and physical dependency There are two subtypes, à 1and à 2 à 1is associated with analgesia, euphoria, and seren- ity, whereas à 2 is associated with respiratory depression, pruritus, prolactin release, dependency, anorexia, and drowsiness
2 κ receptors are present in the limbic and other diencephalic regions, brain stem, and spinal cord They play a role in spinal analgesia, sedation, dyspnea, dependency, dysphoria, and respiratory depression
3 δ receptors are mostly found in the brain However, the effects of these receptors have not been well researched They might potentially cause psy- chomimetic and dysphoric effects à-opioid receptors (MORs) have a vital function in regulating pain relief and are hence the main area of interest in painkiller design research The MORs are a class of receptors that have a role in controlling several physiological activities in the nervous system These receptors mainly influence the perception of pain, but they also have an impact on stress, temperature regulation, breathing, hormonal activity, gastrointestinal function, memory, mood, and motivation Due to their affinity for opioids, these receptors are often known as MORs.
Endogenous ligands
Endogenous ligands, such as β-endorphin, are natural substances that bind to opioid receptors, playing crucial roles in neuro-transmission, pain regulation, and overall bodily function β-endorphin primarily stimulates mu-opioid receptors (MORs) and, to a lesser extent, delta-opioid receptors (DORs) This peptide, derived from proopiomelanocortin, is released by the arcuate nucleus of the hypothalamus in response to stress and exercise, enhancing glucose absorption, inducing pleasure, and alleviating post-exercise discomfort Additionally, β-endorphin exerts a continuous inhibitory effect on gonadotropin-releasing hormone, thereby regulating reproductive functions.
Additional endogenous ligands consist of the enkephalins, which mostly at- tach to DORs and MORs, and the dynorphins, which primarily attach to KORs
Enkephalins are short polypeptides made up of five amino acids, including Met-enkephalin and Leu-enkephalin These pentapeptides are derived from the proenkephalin precursor protein and are primarily found in the amygdala, brainstem, dorsal horn of the spinal cord, adrenal medulla, and various peripheral organs.
Dynorphins, which include dynorphin A (a 17-amino-acid peptide starting with Leu-enkephalin), dynorphin B (or rimorphin), and dynorphin 1-8, are released in key brain regions such as the hippocampus, amygdala, hypothalamus, striatum, and spinal cord These peptides play significant roles in learning, memory, emotional regulation, stress response, and pain perception Additionally, naloxone is effective in counteracting the effects of both naturally occurring and externally administered opioids.
Exogenous ligands
Medications that activate mu-opioid receptors (MORs) are essential for pain relief Among these, codeine and tramadol are considered weak opioids, while oxycodone, morphine, hydromorphone, meperidine, tapentadol, methadone, fentanyl, sufentanil, and remifentanil are classified as strong opioids.
Drugs can have unintended adverse side effects by interacting with proteins not originally targeted The concepts of "biased agonism" and "functional selectivity" describe the ability of a drug to activate specific signaling pathways through the spatial arrangement of molecules in relation to a ligand This phenomenon has garnered significant attention in pharmaceutical research and industry, as it presents a potential therapeutic alternative to traditional opioid pain relievers, such as morphine, which are associated with numerous negative effects like tolerance, dependence, and addiction Consequently, extensive research is focused on developing medications that either specifically bind to their intended protein targets with high affinity (negative design) or enhance the interaction between a drug and its target in living organisms (positive design).
Table 2.1: Examples of opioid drugs, categorized by their origin and type
Heroin Hydrocodone Oxycodone Hydromorphone Oxymorphone
Fentanyl Methadone Meperidine Tramadol Tapentadol Synthetic Opioid Agonist-Antagonists
To summarize, the disparity between two phrases may be succinctly described as follows:
• A biased ligand, or biased agonist, is a kind of ligand that selectively activates one specific receptor transducer pathway in a particular biological system, in comparison to a reference ligand
• An unbiased ligand, or unbiased agonist, is a kind of ligand that activates pathways in a way that cannot be distinguished from the reference ligand
Molecular Docking
Docking and scoring functions
Virtual screening identifies promising drug candidates from extensive compound libraries by targeting molecules that bind to specific proteins or receptors A key technique in this process is molecular docking, which plays a vital role in drug discovery and molecular modeling Ligand-protein docking specifically examines how ligands interact with proteins that have known three-dimensional structures, focusing on their primary binding modes.
Molecular docking generates numerous potential ligand positions based on protein structures, which are evaluated using scoring functions that determine the ligand’s poses These scoring functions are essential as they assess the accuracy of the ligand's positioning within the binding site and predict binding affinity They can be categorized into force-field based, empirical, or knowledge-based types, serving three main purposes: identifying how a ligand binds to a protein by determining the binding mode and site, predicting the binding strength crucial for lead optimization, and facilitating virtual screening to explore extensive ligand databases for potential drug leads targeting specific proteins.
Figure 3.2: Example of Molecular Docking process [35]
Figure 3.3: Four popular scoring functions [30]
Assessing and ranking ligand shapes is crucial for structure-based virtual screening, as the success of binding conformation predictions relies on accurately distinguishing correct poses from incorrect ones The development of reliable scoring functions is essential for quantitative modeling of protein-ligand interactions and binding affinity predictions, aided by free-energy simulation tools However, these complex computations are often unfeasible for large numbers of protein-ligand complexes and lack consistent precision Docking systems' scoring algorithms make certain assumptions and simplifications, failing to fully account for important physical processes like entropic effects in molecular recognition.
Table 3.1: Some popular molecular docking software
Scoring Function Software and Tools
AutoDock Vina Surflex-Dock ChemScore PLANTS X-Score
IT Score DSX SMoG DFIRE KBP
RF Score NNScore DeepChem AtomNet
Deep Learning approaches
Artificial Intelligence (AI) and Machine Learning (ML) are interdisciplinary fields that leverage algorithms and models to perform tasks involving learning, decision-making, and prediction ML, a subset of AI, allows for the development of models through data analysis, eliminating the need for explicit programming Within ML, Deep Learning specifically utilizes multi-layered neural networks designed to replicate the information processing capabilities of the human brain.
Figure 3.4: Example of Neural Network model
Neural networks are a type of supervised machine learning inspired by the human brain, commonly used for tasks such as voice and image recognition A basic neural network consists of an input layer, one or more hidden layers, and an output layer made up of interconnected nodes Each hidden node processes weighted inputs from the previous layer to compute a feature, with outputs flowing through the layers until the output layer generates a classification The choice of network architecture and activation functions significantly influences the network's design, while model parameters are typically fine-tuned to align with a specific training dataset, aiming to minimize errors.
Deep Learning refers to advanced neural networks with multiple layers that can tackle complex tasks, largely thanks to modern graphics card processing power The expressiveness of these networks is determined by their architecture, which defines the number and types of layers that process inputs for classification This architecture can be optimized manually or automatically to enhance performance and reduce overfitting Convolutional Neural Networks (CNN) are a prominent type of neural network tailored for image recognition, analyzing images hierarchically to learn intricate features while maintaining spatial relationships CNNs, exemplified by models like GoogLeNet and Microsoft ResNet, have achieved superior performance in image classification, outperforming human capabilities Their effectiveness in image recognition suggests their potential for learning from various types of spatial data, including protein-ligand structures.
Figure 3.5: Example of CNN model Users label input and output data, computer learns hidden features then returns a best prediction model
Recent studies indicate that structure-based scoring functions utilizing neural networks are on par with empirical scoring methods in virtual screening These scoring functions have also demonstrated their effectiveness in prospective screenings of estrogen receptor ligands In cheminformatics, neural networks have been successfully employed to creatively manipulate 2D chemical structures and design innovative network architectures, serving as viable alternatives to resource-intensive quantum chemical computations.
Figure 3.6: A CNN3D model use for predicting binding affinity of ligand-protein complex
Unlike traditional machine learning methods, a CNN scoring approach eliminates the need for complex feature extraction from structures Instead, it autonomously identifies the key elements essential for precise grading This capability allows for the detection of features that are difficult to represent in simplified models, such as hydrophobic enclosure and surface area-dependent terms, as well as recognizing previously unacknowledged characteristics that current scoring systems may overlook.
Molecular Dynamics Simulation
Equation of Motion
The traditional molecular dynamics method is based on Newton's second law of motion In a system with N particles, each particle i has a mass m_i and a current position r_i The force acting on the i-th particle can be expressed using Newton's equations of motion.
- F i is force acting on particle i th
- m i is mass of particle i th
- a i = d r is acceleration of particle i th tion of the position r i of N particles, denoted as U (r 1 , r 2 , , r N ) Therefore, the
Assuming that the potential energy between particles can be expressed as a func- force exerted on the particle i th may likewise be obtained from the derivative (the gradient):
The MD simulation primarily aims to solve the aforementioned equations 3.3 over a certain time interval.
Integration Algorithm
The system of equations 3.3 addresses the many-body problem, which lacks an analytical solution Therefore, approximations and numerical solutions are more practical approaches Various numerical techniques, particularly those based on finite difference methods, are available, including three prominent algorithms: the Verlet algorithm, the Verlet-velocity algorithm, and the leap frog algorithm.
An algorithm must meet many requirements, including the ability to estimate the correct trajectory over an extended period of time while maintaining accept- able levels of error
1 The algorithm must exhibit time-reversibility
2 The algorithm must have a high level of efficiency in order to execute tasks quickly
3 The algorithm should be straightforward to implement
4 The program must preserve certain macroscopic physical constants
In this thesis, the leap-frog approach is chosen for integrating Newton's equations of motion due to its effectiveness and efficiency compared to other algorithms This method aligns with the principles of ergodic theory in physics and thermodynamics, ensuring accurate estimation of physical values based on system configurations.
In the leap-frog algorithm, the time interval is represented by ∆t, and at the current time t, the position vector of the i-th particle is denoted as r i (t) while its acceleration vector is a i (t) The velocities are initially calculated at the previous time t − ∆t as v i (t − ∆t), following a specific iterative process.
In order to get the total energy at time t, the velocities may be estimated using an approximation method: v (t) = 1 v t + ∆t
Force field model
Current research focuses on large systems that cannot be simulated using quantum mechanics for temporal development, necessitating the use of classical force field models, or molecular mechanics While molecular mechanics is less resource-intensive and can yield comparably accurate results, it has significant limitations, particularly its inability to account for electron mobility, which affects the accuracy of properties dependent on electron distribution.
The general force field model consists of parameters and a potential energy function that regulate processes such as bond stretching, rotation, and angle adjustments Its effectiveness hinges on the accuracy of the Born-Oppenheimer approximation, which simplifies energy calculations based solely on nuclear coordinates Additionally, the model's transferability principle permits the application of parameters from a limited dataset to a wider range of scenarios.
In biological systems, the force field U assumed in equation 3.2 can be ex- i i i Σ Σ pressed as: where:
- E boned : The boned interactions, as known as the total of bond stretching (E bond ), angle bending (E angle ), and bond torsion (E torsion )
- E non−bonded : The non-boned interactions, as known as the total of Van der Waals interaction (E V dw ) and electrostatic interaction (E electro )
All interactions are made explicit below:
1 The bond stretching between two nearest-neighbor atoms, which describes the force exerted by two particles that are linked together, is directly pro- portional to the linear displacement from their equilibrium position r 0, deter- mined by the proportionality constant k b
2 The bending of an angle formed by three consecutive linked atoms The magnitude of the force is directly proportional to the angular displacement mediary particle The angle at equilibrium is denoted as θ 0 and is determined between two particles that are bound together at an angle with a third inter- by the proportionality constant k a
3 The bond torsion between two nearest-neighbor atoms exhibits sinusoidal be- havior, with each term representing the torsion between a pair of particles that are bound together The force is characterized by a proportionality con- stant k ϕ and exhibits sinusoidal variation with the angle ϕ and phase γ It Σ r r Σ i i=1 completes n cycles during one 2π cycle of ϕ
4 The van der Waals interactions are represented by a Lennard-Jones potential
5 The electrostatic potential between two charged particles q i and q j with dis- tance r ij
Analyses methods
Root Mean Square Deviation (RMSD) is a key metric used to measure the difference between a target structure and a reference structure in molecular dynamics simulations This measurement focuses on the temporal evolution of molecular structures compared to their initial states By plotting RMSD as a function of time, researchers can identify significant structural changes in proteins A leveling off or flattening of the RMSD curve indicates that the protein has reached equilibrium, signaling stability in its conformation.
The RMSD at the time t can be expressed as: where:
- N : the number of particles in a group of atom
- m i : the mass of particle i th
- M = Σ N m i : The mass of a group of atom
- r i : the position of particle i th
- r ref : the reference position of particle i th b) Root Mean Square Fluctuation
Root mean square fluctuation (RMSF) is a key metric used to evaluate the flexibility of individual residues during simulations Unlike root mean square deviation (RMSD), which assesses overall structural changes over time, RMSF specifically examines the fluctuations of each residue By plotting RMSF values against residue numbers, researchers can pinpoint which amino acids significantly influence molecular mobility in proteins.
The RMSD of atom i th can be expressed as: where:
- T : the total number of time frame in simulation
- r ref : The time-averaged position of atom i th over trajectory
The Proposed Workflow
The virtual screening process using the Pafnucy model incorporates two key input datasets: protein crystallographic structures obtained from the PDB database and ligands sourced from PubChem catalogs Both datasets are generated as detailed in the following sections.
The molecular docking process initiates with the Pafnucy model, employing Oliceridine (TRV130) as a biased ligand and Morphine-6-Glucuronide (M6G) as an unbiased ligand, guided by biological data Ligands generated through a genetic algorithm with Open Babel software are subsequently utilized in the docking procedure to assess their binding affinity to proteins.
For the final confirmation, the compounds that scored the best in terms of affin- ity were evaluated by Molecular Dynamics simulation using the GROMACS soft- ware [6].
Datasets
Protein structures and preparation
The active mu-opioid receptor (MOR) complexed with G α protein was sourced from the Protein Data Bank (PDB ID: 6DDF) The structure of the MOR-G α complex was elucidated through cryo-electron microscopy (cryo-EM) with a resolution of 3.5 Å This experimental structure featured a MOR comprising 356 amino acid residues, linked to the agonist peptide DAMGO and the nucleotide-binding G i subunit α-1 protein (G α) To address the loss of several residues in the G α structure, the complete G α protein structure was reconstructed using PyMol.
Figure 4.1: Mu Opioid Receptor-Gi Protein Complex [4]
To establish a binding pocket, I pinpointed the central location of the cognate ligand within the complex, which also acts as the center of a cubic box measuring 20 Å This cubic box offers valuable insights into the ligands and atoms in chains that play a direct role in the interaction between ligands and receptors.
Ligand structures preparation
To evaluate the accuracy of the docking software's scoring system, a dataset of 765 opioid molecules, ranging from weak to strong, was sourced from the PubChem database The selection of these drug molecules adhered to Lipinski's rule of five, a qualitative guideline designed to assist chemists in developing orally active compounds.
The rule-of-five is a critical guideline in Pfizer's registration processes, alerting researchers when at least two of its key criteria for therapeutic molecule selection are not satisfied This approach has become increasingly important in the identification of promising compounds for future research.
2 Number of hydrogen bond acceptors < 10
3 Number of hydrogen bond donors < 5
4 Calculated n-octanol-water partition coefficient (ClogP ) < 5
All compounds may be found in Table 6.3.
Generate conformers of biased and unbiased ligands
The goal of conformer generation extends beyond finding a low-energy conformation; it aims to create multiple distinct conformations A genetic algorithm serves as a computational method to identify the globally optimal solution for complex, multi-parameter problems This process involves a population of individuals that evolve over several generations, gradually converging on an optimal solution based on factors such as RMSD diversity or energy levels.
This study focused on generating various poses of Oliceridine (TRV130), a biased agonist, and Morphine-6-glucuronide (M6G), an unbiased agonist The generation utilized genetic algorithms with OpenBabel on the Google Colab Pro platform.
Computational tools
Pafnucy
Pafnucy is an advanced deep neural network specifically designed for structure-based methods, utilizing a single output neuron to predict binding affinity with high accuracy It employs a unique atom description method, ensuring that both proteins and ligands share the same atom types, which acts as a regularization tool to highlight essential interaction properties The model consists of convolutional and dense components interconnected through various layer connections It identifies patterns through filters in the convolutional layer, creating a feature map that illustrates the spatial arrangement of these patterns Pafnucy's architecture offers valuable insights into feature significance, information extraction during the learning process, and the ultimate prediction of binding affinity.
To create a binding pocket, the complex is reduced to a 20Å cubic box centered on the ligand's geometric center The heavy atom locations are then converted into discrete values using a 1 Angstrom precision three-dimensional grid, allowing the input to be represented as a 4D tensor This tensor consists of Cartesian coordinates in the first three dimensions and a feature vector in the final dimension Pafnucy encodes the chemical complex as a 4D tensor, which is processed as a 3D image with multiple color channels Each input location is defined by x, y, and z coordinates and characterized by a vector of 19 attributes, similar to how each pixel in an image is described by its x and y coordinates and intensity values for three primary colors.
The neural network was developed and assessed using protein-ligand complexes sourced from the 2016 version of the PDBbind database This database provides three-dimensional structures of molecular complexes along with their corresponding binding affinities, expressed as pK values (−logK d).
GROMACS
The GROMACS v2020 software was utilized to conduct 500-nanosecond molecular dynamics simulations on selected ligands post-molecular docking, aiming to assess the stability of the ligand-protein complexes The solutes were placed in a cubic container, ensuring a minimum distance of 10 Å from the edges The ligand-protein complexes were prepared using the AMBER99SB-ILDN force field, and to neutralize the system, NaCl was added at a concentration of 150 nM The initial energy minimization was executed using the steepest descent method, with a maximum force constant threshold of 1000 kJ/mol/nm, followed by a 50 ns system equilibration at a controlled temperature.
The simulations were conducted at a temperature of 310 K and a pressure of 1 bar, utilizing the NVT and NPT ensembles with the Nose-Hoover thermostat and Parrinello-Rahman barostat A cubic simulation box with an edge length of 14 nm was employed, ensuring that the complex and its periodic images are at least 3 nm apart to prevent unwanted interactions, such as the electrostatic screening effect With a NaCl concentration of 150 mM, the electrostatic screening length is approximately 7 nm.
The 3 nm distance between proteins in the simulation boxes effectively mitigates finite-size effects caused by long-range electrostatic interactions To manage these interactions, the Particle Mesh Ewald (PME) method is employed with a cutoff length of 1.2 nm, which is also applied to Van der Waals interactions.
Virtual Screening Results
Endogenous ligands
Opioid peptides, naturally produced by the body, play a crucial role in the brain's systems related to motivation, emotion, attachment behavior, stress response, pain management, and food intake regulation Their predicted binding affinities highlight their importance in these essential functions.
Table 5.1: Predicted binding affinity for endogenous ligands
No Endogenous ligands PubChem ID Molecular Formula Predicted pKa
Endomorphins, innate opioid neuropeptides, play a vital role in pain relief, with two main types identified: Endomorphin-1 and Endomorphin-2 These tetrapeptides, consisting of specific amino acid sequences, exhibit strong binding affinity to mu-opioid receptors (MORs), leading to the suppression of neuronal activity Endomorphin-1 is predominantly found in the brain and upper brainstem, while Endomorphin-2 is concentrated in the spinal cord and lower brainstem, highlighting their significant presence in both the central and peripheral nervous systems.
Dynorphins, a group of opioid peptides derived from the precursor protein prodynorphin, primarily act through the kappa-opioid receptor (KOR) while also showing some affinity for mu-opioid receptors (MOR) and delta-opioid receptors (DOR).
The results illustrate the good binding affinity between endogeneous ligands and MOR.
Non-opioid ligands
Table 5.2: Predicted binding affinity for non-opioid ligands
No Non-opioid ligands PubChem ID Molecular Formula Predicted pKa
3 Gamma-aminobutyric acid (GABA) 119 C 4 H 9 NO 2 4.4248505
Those above ligands can be found in:
• flavoring agents such as vanilin used in bakery (benzaldehyde)
• dyes and rubber (aniline, 2-Chloroaniline)
GABA, an endogenous amino acid, is crucial for the proper functioning of the central nervous system, helping to regulate emotions linked to stress, anxiety, and fear, and it is not classified as an opioid.
S-3-Isobutylpiperazine-2,5-dione binds to the active site of human DNA polymerase alpha, effectively inhibiting DNA production When used in combination with mithramycin, it demonstrates synergistic effects that further suppress leukemia cell proliferation.
• Aspirin(Paracetamol) is a popular painkiller in use, however, it is a non-opioid analgesic
• 2-((Dimethylamino)methyl)cyclohexanone is an industrial chemicals
• (+)-[3h]Pentazocine binds to σ-opiod receptor
Pafnucy’s CNN model accurately identified the chemicals that had weak or non-binding affinity for MOR.
Opioid Drugs
Figure 5.1: Predicted Binding Affinity of several opioid drugs with MOR
This image illustrates the binding affinity of three categories of opioid drugs:
• Semi-synthetic opioids: Heroin, Oxycodone, Hydrocodone, Hydromorphone, Oxymorphone
• Synthetic opioids: Fentanyl, Methadone, Tramadol, Buprenophine, Meperi- dine
The binding affinity of opioid medications to the mu-opioid receptor is illustrated in Figure 5.1, with pK d values typically starting at 5.5 Many drugs utilized in virtual screening exhibit a binding affinity exceeding this threshold Consequently, employing a CNN 3D model for virtual screening proves to be highly effective in identifying potential candidates for mu-opioid receptor interactions.
However, Fentanyl is a very potent opioid, with a strength that is about 50-
Recent virtual screening data suggest that Fentanyl's potency compared to Morphine remains inconclusive This uncertainty arises from the use of original poses from PubChem during the screening process, which did not consider the optimal docking conformations that could influence potency assessments.
Pose Docking prediction
TRV130 with MOR
Table 5.3: Docking results of five best poses and five worst poses of Oriceridine (TRV130)
Rank Pose label Binding Affinity Rank Pose label Binding Affinity
Figure 5.3: 5 best poses of TRV130 Figure 5.4: 5 worst poses of TRV130
The figures illustrate various postures of TRV130 within the 6DDF pocket, highlighting that the top two highest scoring poses interact with the residue TYR148R in the transmembrane protein In poses 47, 62, 36, and 42, a nitrogen atom in the ring forms a hydrogen bond with the hydroxyl group of residue TYR148R.
TRV130 tance of about 3.1 A˚ In pose 30, there is an aromatic interaction with TYR148R Moreover, there is a specific interaction occurring between the Sulfur atom at pose
30 of the ring and the NH 2 group in ARG211R, with a distance of 3.2 A˚
The oxygen atom at positions 63, 75, and 92 forms a bond with the NH2 group of the ARG221R residue, while poses 77 and 80 feature a nitrogen atom interacting with a hydroxyl group in TYR148R The measured distances for the N-H-O and aromatic interactions are 3.3 Å and 4.7 Å, respectively.
M6G with MOR
Table 5.4: Docking results of three best poses and three worst poses of
Rank Pose label Binding Affinity Rank Pose label Binding Affinity
Figure 5.15: Three poses of M6G with the highest affinity
Figure 5.16: Three poses of M6G with the lowest affinity
Figure 5.19: 2D diagram of pose 2 M6G Figure 5.20: 3D diagram of pose 2 M6G
Pose 18, 1, and 20 demonstrate four interactions that significantly strengthen the binding of M6G to the mu-opioid receptor (MOR) Notably, these poses form two bonds with TYR148R through its aromatic ring However, the interactions associated with poses 16 and 21 cannot be identified using PoseView.
The protein complexes, including pose 47 of TRV130 and pose 18 of M6G, are simulated in Molecular Dynamics, respectively.
Molecular Dynamics simulation results
RMSD and RMSF metrics are derived from molecular dynamics simulations conducted on our systems, focusing on two distinct ligand conformations: TRV130 (blue) and M6G (orange), illustrated in Figures 5.21 and 5.22.
Figure 5.21: RMSD of the backbone of the à-Opioid receptor
We evaluate the stability of our complex by measuring the root mean square deviation (RMSD) of the protein structures from their original configurations, as illustrated in Figure 5.21 The RMSD of the receptor shows a gradual increase, attributed to the initial frozen structure derived from X-ray studies, indicating that the protein undergoes structural changes during the simulation due to thermal expansion at 310 K Notably, the RMSD of the MOR backbone remains stable for 120 nanoseconds Our current research emphasizes the binding characteristics of àOR with potential ligands, with a primary focus on MOR stability The observed RMSD values are considered acceptable, reflecting the overall stability within the system.
Figure 5.22: RMSF of the backbone of the à-Opioid receptor
To assess the stability of the mu-opioid receptor (MOR), we calculated the root mean square fluctuation (RMSF) for each residue, revealing a mean deviation of the backbone ranging from 0.05 to 0.35 nm MOR, a member of the G protein-coupled receptor (GPCR) family, features seven transmembrane helices along with three intracellular loops, three extracellular loops, an extracellular N-terminal, and an intracellular C-terminal domain The N- and C-terminals exhibit minimal restrictions, leading to increased fluctuations Our model's variation of MOR is consistent with its structural characteristics, demonstrating that seven transmembrane proteins show limited variation due to their confinement within the membrane, unlike other components in aqueous environments that experience significantly greater fluctuations.
This article reviews the validation of models in small molecule screening and deep learning for de novo drug discovery It highlights that while deep learning shows potential in drug design, significant shortcomings exist in model construction and evaluation, particularly in the scoring system that fails to accurately represent chemical bonding interactions The predictive algorithms struggle to differentiate between optimal and suboptimal binding poses, and their limitations arise from the inability to incorporate flexible ligands or structures, which are crucial for realistic biological binding processes Despite these challenges, deep learning can provide scientists with rapid insights into potential candidates for various proteins, but further validation through Molecular Dynamics simulations and biological data is essential to ensure reliability in drug design.
This study does not assess the effectiveness of medications targeting à-opioid receptors Instead, it primarily examines opioid drugs employed in molecular docking, frequently prescribed or used in medical procedures.
Recent advancements in science and technology have led to innovative molecular docking methods that utilize neural network models These models enhance the understanding of complex movements during docking operations, ultimately improving the reliability and efficiency of molecular docking in drug discovery.
[1] 7.5: Tertiary structure of proteins https://chem.libretexts.org/
Bookshelves/Introductory_Chemistry/Introduction_ to_Organic_and_Biochemistry_%28Malik%29/07%3A_
Proteins/7.05%3A_Tertiary_structure_of_proteins
[2] Amino Acid Structure - Definition, Structure, Basicity of amino acid with examples https://byjus.com/chemistry/ amino-acid-structure/
[3] Levels of Protein Structure https:
//www.creative-biostructure.com/ levels-of-protein-structure.htm
[4] PDB ID: 6DDF https://doi.org/10.2210/pdb6ddf/pdb [Ac- cessed 19-06-2024]
[5] Protein Structure - Primary, Secondary, Tertiary, Qua- ternary https://www.geeksforgeeks.org/ protein-structure-primary-secondary-tertiary-quaternary/
[6] Mark Abraham, Andrey Alekseenko, Vladimir Basov, Cathrine Bergh,
Eliane Briand, Ania Brown, Mahesh Doijade, Giacomo Fiorin, Stefan Fleis- chmann, Sergey Gorelov, Gilles Gouaillardet, Alan Grey, M Eric Irrgang,
Farzaneh Jalalypour, Joe Jordan, Carsten Kutzner, Justin A Lemkul, Mag- nus Lundborg, Pascal Merz, Vedran Miletic, Dmitry Morozov, Julien Nabet,
Szilard Pall, Andrea Pasquadibisceglie, Michele Pellegrino, Hubert Santuz,
Roland Schulz, Tatiana Shugaeva, Alexey Shvetsov, Alessandra Villa, Sebas- tian Wingbermuehle, Berk Hess, and Erik Lindahl Gromacs 2024.2 manual, May 2024
[7] Maria Batool, Bilal Ahmad, and Sangdun Choi A structure-based drug dis- covery paradigm Int J Mol Sci., 20(11):2783, June 2019
[8] Ekaba Bisong Google Colaboratory, pages 59–64 Apress, Berkeley, CA,
A study by Durrant et al (2015) published in the Journal of Chemical Information and Modeling highlights the use of neural-network scoring functions to identify novel ligands for estrogen receptors This research emphasizes the potential of advanced computational methods in discovering new compounds that can interact with important biological targets.
[10] Abdelfattah Faouzi, Balazs R Varga, and Susruta Majumdar Biased opioid ligands Molecules, 25(18), 2020
[11] Rafael Franco, David Aguinaga, Jasmina Jime´nez, Jaume Lillo, Eva Mart´ınez-Pinilla, and Gemma Navarro Biased receptor functionality ver- sus biased agonism in g-protein-coupled receptors Biomolecular Concepts,
In their November 2021 article published in Brief Bioinformatics, Thomas Gaudelet and colleagues explore the application of graph machine learning in drug discovery and development The study emphasizes the potential of leveraging advanced graph-based algorithms to enhance the efficiency and effectiveness of identifying new therapeutic compounds By integrating machine learning techniques with graph representations of molecular structures, the authors highlight significant advancements in predictive modeling and data analysis within the pharmaceutical industry.
In 2021, Zahra Sadat Hashemi and colleagues conducted a study on the design and optimization of interfering peptides aimed at disrupting protein–protein interactions through in silico approaches Their research highlights the significance of computational methods in peptide design, contributing to advancements in biochemical and pharmaceutical applications.
[14] John Jumper, Richard Evans, Alexander Pritzel, Tim Green, Michael Fig- urnov, Olaf Ronneberger, Kathryn Tunyasuvunakool, Russ Bates, Augustin
In August 2021, Zˇ ´ıdek et al published a groundbreaking study in Nature, demonstrating the high accuracy of protein structure prediction using AlphaFold The research, involving prominent scientists like Anna Potapenko, Alex Bridgland, and Demis Hassabis, highlights the significant advancements in computational biology, paving the way for enhanced understanding of protein functions and interactions This study marks a pivotal moment in the field, showcasing the potential of artificial intelligence to revolutionize biological research and therapeutic development.
[15] Sunghwan Kim, Jie Chen, Tiejun Cheng, Asta Gindulyte, Jia He, Siqian
He, Qingliang Li, Benjamin A Shoemaker, Paul A Thiessen, Bo Yu, Leonid Zaslavsky, Jian Zhang, and Evan E Bolton PubChem 2023 update Nucleic Acids Research, 51(D1):D1373–D1380, 10 2022
[16] Joshua Li, Yuhong Ning, Warren Hedley, Brian Saunders, Yongsheng Chen, Nicole Tindill, Timo Hannay, and Shankar Subramaniam The molecule pages database Nature, 420(6916):716–717, Dec 2002
[17] Zhihai Liu, Minyi Su, Li Han, Jie Liu, Qifan Yang, Yan Li, and Renxiao Wang Forging the basis for developing protein-ligand interaction scoring functions Acc Chem Res., 50(2):302–309, February 2017
[18] Kenneth T Luu, Eugenia Kraynov, Bing Kuang, Paolo Vicini, and Wei-Zhu Zhong Modeling, simulation, and translation framework for the preclinical development of monoclonal antibodies AAPS J., 15(2):551–558, April 2013
[19] Noel M O’Boyle, Michael Banck, Craig A James, Chris Morley, Tim Van- dermeersch, and Geoffrey R Hutchison Open babel: An open chemical toolbox Journal of Cheminformatics, 3(1):33, Oct 2011
[20] World Health Organization Opioid overdose https://www.who.int/ news-room/fact-sheets/detail/opioid-overdose [Ac- cessed 19-06-2024]
[21] Kinam Park and Andrew Otte Prevention of opioid abuse and treatment of opioid addiction: Current status and future possibilities Annual Review of Biomedical Engineering, 21(Volume 21, 2019):61–84, 2019
[22] Kirsten M Raehal, Cullen L Schmid, Chad E Groer, and Laura M Bohn Functional selectivity at the à-opioid receptor: Implications for understand- ing opioid analgesia and tolerance Pharmacological Reviews, 63(4):1001–
[23] Matthew Ragoza, Joshua Hochuli, Elisa Idrobo, Jocelyn Sunseri, and David Ryan Koes Protein–ligand scoring with convolutional neural net- works Journal of Chemical Information and Modeling, 57(4):942–957,
[24] Cle´mence Re´ da, Emilie Kaufmann, and Andre´e Delahaye-Duriez Machine learning applications in drug development Comput Struct Biotechnol J.,
[25] Sadam D V Satyanarayana, M S R Krishna, Pindi Pavan Kumar, and Sirisha Jeereddy In silico structural homology modeling of nif a protein of rhizobial strains in selective legume plants J Genet Eng Biotechnol., 16(2):731–737, December 2018
[26] Schro¨dinger, LLC The PyMOL molecular graphics system, version 1.8 November 2015
[27] Phil Skolnick The opioid epidemic: Crisis and solutions Annual Review of
[28] Sebastian Stabinger, David Peer, Justus Piater, and Antonio Rodr´ıguez- Sa´nchez Evaluating the progress of deep learning for visual relational con- cepts J Vis., 21(11):8, October 2021
[29] M Alexander Stanczyk and Ram Kandasamy Biased agonism: the quest for the analgesic holy grail PAIN Reports, 3(3), 2018
[30] Francesca Stanzione, Ilenia Giangreco, and Jason C Cole Chapter four - use of molecular docking computational tools in drug discovery volume 60 of Progress in Medicinal Chemistry, pages 273–343 Elsevier, 2021
[31] Marta M Stepniewska-Dziubinska, Piotr Zielenkiewicz, and Pawel Siedlecki Development and evaluation of a deep learning model for protein–ligand binding affinity prediction Bioinformatics, 34(21):3666–3674, 05 2018
[32] Katrin Stierand and Matthias Rarey Drawing the pdb: Protein-ligand com- plexes in two dimensions ACS Medicinal Chemistry Letters, 1(9):540–545,
[33] Duxin Sun, Wei Gao, Hongxiang Hu, and Simon Zhou Why 90% of clin- ical drug development fails and how to improve it? Acta Pharm Sin B.,
The article by Tohru Takebe, Ryoka Imai, and Shunsuke Ono discusses the current state of drug discovery and development in the United States, highlighting the significant impact of collaboration between academia and industry It emphasizes how these partnerships enhance the innovation process in pharmaceuticals, ultimately advancing clinical and translational science The research, published in November 2018, provides insights into the evolving landscape of drug development and the essential role of collaborative efforts in driving progress.
[35] Yair Tenorio, Alejandra Hernandez-Santoyo, Victor Altuzar, Hector Vivanco-Cid, and Claudia Mendoza-Barrera Protein-Protein and Protein- Ligand Docking, page 187 05 2013
[36] Manthena VS Varma, Omathanu P Perumal, and Ramesh Panchagnula Functional role of p-glycoprotein in limiting peroral drug absorption: opti- mizing drug delivery Current Opinion in Chemical Biology, 10(4):367–373,
[37] Olivier J Wouters, Martin McKee, and Jeroen Luyten Estimated research and development investment needed to bring a new medicine to market, 2009-2018 JAMA, 323(9):844–853, March 2020
All results of virtual screening process can be found on page 56, 57 and 58
The predicted binding affinity of M6G with MOR can be found in Table 6.1 on page 54
The predicted binding affinity of TRV130 with MOR can be found in Table 6.2 on page 55
Table 6.1: Docking results of 24 poses of Morphine-6-Glucuronide (M6G)
Rank Pose label Binding Affinity Rank Pose label Binding Affinity
Table 6.2: Docking results of 100 poses of Oliceridine (TRV130)
Rank Pose label Binding Affinity Rank Pose label Binding Affinity
Figure 6.1: Results of Virtual Screening process
Figure 6.2: Results of Virtual Screening process
Family Name of medicine PubChemID Predicted pK
Remifentanil 1-(Methoxycarbonyl-ethyl)-4-[(2-methyl-3-oxo
-pentanoyl)-phenyl-amino]-piperidine -4-carboxylic
Morphine Morphine 3-?-D-Glucuronide Methyl Ester 163285670 6.687029 Oxymorphone Oxymorphone 4-nitrophenylhydrazone 9576842 6.685487
Morphine US9233167, a-6-mPEG2-O-Morphine 124037291 6.6097345 Fentanyl (+/-)-cis-3-Carbomethoxy fentanyl 15358486 6.6046505
-N,N,14-trimethyl-15-oxatetracyclo[10.2.1.05,14.08,13] pentadeca-3,8(13),9,11-tetraen-6-amine
(2R,3S,3aR,5aR,6R,11bR,11cS)-3a-methoxy-3,14-dimethyl -2-phenyl-2,3,3a,6,7,11c-hexahydro-1H-6,11b-(epiminoethano) -3,5a-methanonaphtho[2,1-g]indol-10-ol
Morphine Morphine-6-ol, 14-bromo-, trimethyl ether 628239 6.5151196 M3G Morphine 3-?-D-Glucuronide Methyl Ester 163285670 6.5144186
Morphine Morphinan-3,6-diol, 7,8-didehydro-4,5-epoxy-17-methyl-
Oxymorphone Oxymorphone fumarate methyl ester 6442029 6.4902678 Fentanyl 3-Furancarboxamide fentanyl phenyl-d5 137699978 6.487301
Morphine 6-Dehydro-N-desmethyl-N-(phenylmethoxy)carbonyl
Buprenophine (-)-3-Methoxy Butorphanol 6-Ethylene Ketal 71750031 6.4622087 Butorphanol (-)-3-Methoxy Butorphanol 6-Ethylene Ketal 71750031 6.4622087
Codeine US9233167, a-6-mPEG2-O-Codeine 124037298 6.3975973 Morphine 3-trifluoromethanesulfonyl morphine 67513718 6.3966417
Codeine US10512644, Compound alpha-6-mPEG4-O-Codeine 154732853 6.3646936 Oxymorphone 14-Hydroxydihydromorphinone hydrazone 9576841 6.364462
Fentanyl para-Fluoro Valeryl fentanyl 137700047 6.3554697
Fentanyl para-Fluoro-3-furanyl fentanyl 137700041 6.351679
Carfentanil N-Despropionyl N-Acetyl Carfentanil Methyl Ester 165361602 6.3275795
Fentanyl 4-Fluoro cyclopropyl benzyl fentanyl 154572819 6.3168535 Alfentanil
N-(1-(2-(4-Ethyl-5-oxo-4,5-dihydro-1H-tetrazol-1-yl)ethyl) -4-((propanoyloxy)methyl)piperidin-4-yl)
Morphine Morphine, 7-hydroxy-6,6-dimethoxy-3-O-methyl- 632235 6.304097
Morphine Compound alpha-6-mPEG2-O-Morphine 53327324 6.254257 Morphine 4-[3-(4-Benzylpiperazin-1-yl)-2,5-dioxopyrrolidin-1-yl] benzoic acid
Morphine US9233167, a-6-mPEG3-O-Morphine 124037292 6.218809 Hydromorphone Hydromorphone, pentafluoropropionate 6426024 6.218236 Codeine US10512644, Compound alpha-6-mPEG1-O-Codeine 154732850 6.2162714
Fentanyl p-Fluoro methoxyacetyl fentanyl 137699935 6.202382 Remifentanil methyl 1-benzyl-4-
Fentanyl para-Fluoro cyclopropyl fentanyl 137700017 6.201686 Fentanyl trans-3-Carbomethoxy fentanyl 15358487 6.2015715 Fentanyl para-Fluorobutyryl fentanyl-d7 137699876 6.20142 Fentanyl Para-fluoro benzyl fentanyl 10545121 6.199111
(2-phenylethyl)-4-piperidinyl)-N-phenyl- 5745882 6.1781197 Fentanyl Methoxyacetyl fentanyl-d5 137699967 6.178056
Alfentanil N-(1-(2-Hydroxyethyl)-4-(methoxymethyl) piperidin-4-yl)-N-phenylpropionamide 145996686 6.176528
Codeine Morphinan-4-ol, 3-methoxy-17-methyl- 591302 6.1667757
Fentanyl N-(1-Benzylpiperidin-4-yl)-N-phenylbenzamide 10044685 6.1608686 Tramadol Desmethyl tramadol O-b-D-glucuronide 46781195 6.1605415
Morphine (4R,4aR,7S,7aR,12bS)-3-methyl-2,3,4,4a,7,7a-hexahydro
Tramadol rac N-Benzyl-N-desmethyl Tramadol-d3 71313807 6.146579
Fentanyl para-Chloro cyclopropyl fentanyl 137700088 6.1460323
Fentanyl para-Methyl Cyclopropyl fentanyl 137700045 6.1254463
Morphine 2-n-pentyloxy-2phenyl-4-methyl-morphine 129695836 6.1251907
Carfentanil N-(Desoxopropyl) N-(3-Hydroxy-1-oxopropyl) Carfentanil 155520187 6.114956
Fentanyl para-hydroxy Butyryl fentanyl 156613767 6.114506
Fentanyl beta-Methyl acetyl fentanyl 615726 6.113328
Remifentanil Remifentanil bis ethyl ester 165361620 6.106848 Morphine Morphinan-3-ol, 6,7-didehydro-4,5-epoxy-6,17
Fentanyl p-Chloro methoxyacetyl fentanyl 13653630 6.101284 Buprenophine (-)-3-Methoxy Butorphanol 6-Ethylene Ketal 138395412 6.09809 Butorphanol (-)-3-Methoxy Butorphanol 6-Ethylene Ketal 138395412 6.09809
Codeine US10512644, Compound alpha-6-mPEG3-O-Codeine 154732852 6.0897074
Fentanyl Benzofuranyl-fentanyl 165361426 6.081976 Remifentanil Remifentanil hydrochloride impurity H 135391032 6.0819483
Morphine US10512644, Compound alpha-6-mPEG1-O-Morphine 53327323 6.069989 Dihydrocodeine 14-hydroxy Dihydrocodeine N-oxide 165362010 6.06989
Fentanyl Tetrahydrofuranyl fentanyl ring-opened-alcohol 157010657 6.0607853
Tramadol O-Desmethyl Tramadol beta-D-Glucuronide 129627216 6.0561905 Fentanyl 4’-Hydroxy-tetrahydrofuranylfentanyl 157010641 6.052417
Hydromorphone Hydromorphone-3-glucoside 131770048 6.039049 Fentanyl Valeryl fentanyl carboxy metabolite 56672265 6.0389614
Pentazocine Cortisone 21-cyclopentanepropionate 22879767 6.0366116 Fentanyl o-Methyl-methoxyacetylfentanyl 137700051 6.034708 Fentanyl para-Methoxy-Butyrylfentanyl-d7 137699900 6.034254
Remifentanil p-Fluoro-remifentanil ethyl ester” 165361614 6.031395
Fentanyl 2-Hydroxy-N-phenyl-N-[1-(2-phenylethyl)-4-piperidinyl] propanamide; ?-1-Hydroxyfentanyl
Fentanyl N-methyl meta-methyl Phenyl fentanyl 156589019 6.019891
Buprenophine Morphinan, 14-hydroxy-3-methoxy-17-cyclobutylmethyl- 625439 6.0188036 Butorphanol Morphinan, 14-hydroxy-3-methoxy-17-cyclobutylmethyl- 625439 6.0188036
Ketobemidone Ketobemidone,(3,3-diMe-butryl)ester 23276586 6.017541 Remifentanil N’-Despropionyl-N’-acetyl remifentanil 14987174 6.016631
Tramadol rac N,O-Didesmethyl Tramadol O-Sulfate 71316033 6.0153513
Morphine Morphine 6-beta-D-glucuronide hydrate 9868916 6.0025406
Fentanyl 1-Benzyl-4-(phenylamino)piperidine-4-carbonitrile 70412 5.971961
Hydromorphone 6alpha-Hydroxy-hydromorphone 131769954 5.965008 Fentanyl methyl(1-phenethylpiperidin-4-yl)
Morphine 3-(p-chlorobenzyl)-morphine 129780026 5.9456196 Fentanyl Despropionyl p-Fluoro Fentanyl 21812144 5.9445887 Tramadol O-Desmethyl Tramadol b-D-Glucuronide 96360331 5.941572
Remifentanil Methyl 3-(4-cyano-4-(phenylamino)piperidin-1-yl) propanoate 24780059 5.9399548
Morphine Compound alpha-6-mPEG3-O-Morphine 53327325 5.935976
Carfentanil Carfentanil metabolite M2 12298033 5.918334 Fentanyl Desethylbenzene Fentanyl 4-Methoxymethyl
Fentanyl N-Methyl ortho-methyl Phenyl fentanyl 165361531 5.9046316
Hydromorphone Nordihydroisomorphine 131769955 5.8465605 Fentanyl 1-Benzyl-N-phenylpiperidin-4-amine 70865 5.843545
Hydromorphone Hydromorphone 3-glucuronide 10479418 5.841093 Loperamide 4-(4-chlorophenyl)piperidin-4-ol” 38282 5.841057
Tramadol (-)-N,N-Bisdesmethyl Tramadol 40424854 5.8191643 Fentanyl Ethyl Despropionyl fentanyl 156346345 5.8160105
Morphine 7,8-Dihydro-14-hydroxymorphine 627484 5.7853327 Norpropoxyphene Norpropoxyphene amide 12372434 5.781472
Morphine Morphinan-3,6-diol, 7,8-didehydro-4,5-epoxy-2-
BMS-986122 2-(3-bromo-4-methoxyphenyl)-3-(4-chlorophenyl) sulfonyl-1,3-thiazolidine 4644453 5.720524
Beta-Endorphin tyrosyl-glycyl-glycine 123715 5.709033
Methadone 1,5-Dimethyl-3,3-diphenyl-2-ethylidenepyrrolidine 5352621 5.567316 Meperidine 4-Piperidinecarboxylic acid, 1-methyl-4-phenyl-,methyl ester 551426 5.5633163 Fentanyl omega-1 Hydroxypropionyl nor-fentanyl 154113448 5.558012