Thoth's progress is powered by extensive global partnerships, spanning academic and industrial labs across Canada and internationally. Below, we spotlight notable scientific accomplishments arising from these collaborations. It's worth noting that a significant share of these influential publications was spearheaded by our dedicated team members, underscoring their leadership and our steadfast dedication to advancing computational drug discovery through our collective expertise.
Discovery of allosteric SHP2 inhibitors
SHP2, also known as Src homology-2 domain-containing protein tyrosine phosphatase-2, stands as a pivotal cytoplasmic enzyme whose genetic blueprint is encoded by the PTPN11 gene. Its role in the intricate choreography of cell growth and differentiation cannot be overstated. SHP2's significance extends further as it has been identified as an oncoprotein, implicated in developmental disorders and various cancer types, including gastric, leukemia, and breast cancer. Consequently, it has emerged as a prime target for therapeutic intervention, drawing the spotlight of current research endeavors.
In particular, the focus of these investigations has shifted towards the development of allosteric inhibitors of SHP2. Allosteric inhibitors offer distinct advantages, boasting enhanced selectivity and pharmacological appeal when compared to competitive catalytic inhibitors, which directly target SHP2's catalytic site. However, the pursuit of novel allosteric inhibitor scaffolds aimed at SHP2 continues unabated, driven by the aspiration to craft compounds that exhibit superior selectivity, bolstered cell permeability, and enhanced bioavailability.
This study represents a concerted effort towards realizing this noble ambition. Employing an array of computational tools, the research embarked on an exhaustive screening process, sifting through a vast library of over six million compounds. These computational methodologies, comprising molecular docking studies, structure-based drug design, and simulation techniques, were meticulously employed to scrutinize the potential of these molecules to bind to SHP2 at allosteric sites, thereby influencing its activity.
The research outcome unveiled a select group of top-ranked hits, identified through in-silico screening, which were subsequently subjected to validation using protein thermal shift and biolayer interferometry assays. These rigorous assays confirmed the potency of three specific compounds. Further, kinetic binding assays were employed to quantify the binding affinities of these top-ranked compounds, demonstrating their robust nanomolar affinity for SHP2.
In summation, this study represents a significant stride towards the identification and design of a new generation of allosteric SHP2 inhibitors. The compounds highlighted herein, along with the computational workflow employed, underscore an effective strategy for the discovery of improved inhibitors, promising to usher in innovative therapeutic avenues for SHP2-related conditions and malignancies, offering hope to those afflicted by them.
January 2023
Computers in Biology and Medicine
Identifying novel allosteric sites in the SARS-Cov-2 polymerase assembly
The replication of the SARS-CoV-2 genome is a crucial step in the virus's life cycle, and recent research has shown that inhibiting the replicase machinery of the virus holds promise in combating it. However, several aspects related to the structure, function, and dynamics of the CoV-2 polymerase remain unexplored. These include understanding the dynamic behavior of different polymerase subdomains, investigating hydrogen bond networks at the active site and template entry point in the presence of water, examining how nucleotides bind at the active site, identifying positions in the nascent RNA strand where nucleotide substitutions are acceptable, pinpointing potential allosteric sites within the polymerase, and studying their correlated dynamics in relation to the catalytic site.
To address these knowledge gaps, a comprehensive approach was taken in this study. Cutting-edge modeling tools were combined with recently resolved SARS-CoV-2 cryo-EM polymerase structures to provide insights into these critical aspects. The findings of this research offer a detailed analysis of the hydrogen bond networks at various locations within the polymerase structure. Additionally, the study suggests possible nucleotide substitutions that the polymerase complex can tolerate, which is essential for understanding the flexibility of the replication process.
Furthermore, the study identifies three allosteric sites within the NSP12 RdRp (RNA-dependent RNA polymerase), which are potential targets for small molecule inhibitors. These allosteric sites are referred to as "druggable" because they are amendable to therapeutic intervention. Notably, the research demonstrates that the dynamics within one of these newly discovered allosteric sites are correlated with the activity of the active site. This finding suggests that targeting this specific site could have a significant impact on the catalytic activity of the SARS-CoV-2 polymerase, offering a potential avenue for the development of novel antiviral drugs.
In summary, this study provides a comprehensive exploration of critical aspects of the SARS-CoV-2 polymerase, shedding light on its structure, function, and dynamics. The findings contribute valuable insights that could guide the development of effective therapeutic strategies for combatting the virus by targeting the polymerase replicase machinery.
December 2022
Journal of Biomolecular Structure and Dynamics
Predicting cardiotoxicity using ToXTree: descriptor-based machine learning models.
The study focuses on the development of predictive models for assessing the liability of drugs to affect two crucial cardiovascular channels: the human ether-à-go-go-related gene (hERG) voltage-gated potassium channel and the Nav1.5 voltage-gated sodium channel. Blockade of these channels by drugs can lead to severe cardiovascular complications, making it essential to evaluate their potential impact during drug development.
Two robust predictive models were created using machine learning techniques. The first model, ToxTree-hERG Classifier, employed Random Forest models and was trained on a substantial dataset consisting of 8,380 distinct molecular compounds. This model predicted the potency of compounds in relation to hERG channel blockade at three different potency cut-offs (1µM, 10µM, and 30µM). The results indicated that the hERG model achieved high accuracy in multiclass classification (Q4 = 74.5%) and binary classification (Q2 = 93.2%), with excellent sensitivity (98.7%), specificity (75%), Matthews Correlation Coefficient (MCC = 80.3%), and Correct Classification Rate (CCR = 86.8%) on an external test set of 499 compounds. This model outperformed many existing models and tools in predicting hERG channel liability.
The second model, ToxTree-Nav1.5 Classifier, employed kernelized Support Vector Machine (SVM) models and was trained using a manually curated dataset of 1,550 unique compounds from publicly available bioactivity databases (ChEMBL and PubChem). This model aimed to predict the liability of compounds to affect the Nav1.5 channel. The Nav1.5 model demonstrated a multiclass accuracy (Q4 = 74.9%) and binary classification performance (Q2 = 86.7%) on an external test set of 173 unique compounds. It exhibited strong MCC (71.2%) and F1 score (89.7%). Notably, this study presents the first predictive model for Nav1.5 liability.
In conclusion, these newly developed predictive models offer valuable tools for assessing the potential cardiovascular risks associated with drugs by predicting their effects on hERG and Nav1.5 channels. The models demonstrated high accuracy and outperformed existing tools, contributing to the advancement of drug safety assessment and the reduction of cardiotoxicity-related issues during drug development.
May 2022
Research Square
Classifying GPCR-targeting Compounds UsingGPCR_LigandClassify.py; a Robust Machine Learning Tool
This study focuses on the development of ligand-based machine learning models for drug repurposing against G-Protein Coupled Receptors (GPCRs). GPCRs are a diverse family of receptors with significant implications in drug discovery and therapeutics. The research involved the construction of these models using a substantial dataset comprising over 500,000 data points, which encompassed molecular association information for more than 160,000 unique ligands and over 250 GPCRs. These data points were sourced from the GPCR-Ligand Association (GLASS) database.
To describe the input molecules comprehensively, the study employed various molecular featurization methods. Multiple supervised machine learning (ML) algorithms were developed, tested, and compared for their predictive accuracy. Notably, the study revealed that ensemble decision trees and gradient boosted trees, combined with molecular fingerprinting, achieved classification accuracy levels approaching those of deep neural networks (DNNs)-based algorithms. On a test dataset, these models demonstrated an impressive performance, achieving an approximate 90% classification accuracy.
Furthermore, the research provides concrete examples of how these models successfully identified connections between known drugs from the DrugBank database and specific members of the GPCR receptor family. These findings align closely with experimental observations previously reported in the scientific literature, validating the effectiveness of the developed ML models.
In conclusion, the study offers a robust framework for the application of machine learning in drug repurposing efforts against GPCRs. By leveraging a vast dataset and employing diverse featurization methods and ML algorithms, the research has produced models that exhibit high classification accuracy. These models hold great potential for identifying novel therapeutic uses for existing drugs and can contribute to the ongoing efforts in computational drug discovery and drug repurposing, thereby accelerating the development of new treatments and therapies.
May 2021
Scientific Reports
Exploring the impact of selective calcium channel blockers on ion transit in human Cav1.2: A computational insite
Selective calcium channel antagonists play a pivotal role in the management of cardiovascular disorders, including hypertension, angina pectoris, and cardiac arrhythmias. These antagonists fall into two primary categories: 1,4-dihydropyridine (1,4-DHPs) and non-DHPs. The non-DHPs category further divides into phenylalkylamines (PAAs) and benzothiazepines (BZTs) derivatives. While these drugs are known for their effectiveness, the precise structural mechanisms underlying their actions remain somewhat elusive.
This study employs a near-open confirmation (NOC) model of the Cav1.2 cardiac ion channel to investigate how these antagonists bind within the pore domain of the channel and affect the fenestration of the pore-forming domains. The impact of calcium ion permeation in the presence of these drug molecules is assessed through steered molecular dynamics (SMD) simulations.
The findings of this research shed light on the blocking activity of these antagonists within the Cav1.2 channel. Nicardipine, a derivative of 1,4-DHPs, exhibits robust blocking activity, necessitating a force exceeding 2500 pN to move calcium ions towards the channel's pore when the compound is present. Similarly, verapamil, a PAA derivative, demonstrates significant blocking activity, requiring almost 2300 pN of force. In contrast, Diltiazem, a BZT derivative, exhibits the least blocking activity.
These results provide insights into the structural basis and binding mechanisms of 1,4-DHPs, PAAs, and BZTs within the distinct sites of the Cav1.2 channel. They offer a comprehensive understanding of how these antagonists modulate the channel's function, thereby contributing to our knowledge of their mechanisms of action.
In summary, this study enhances our comprehension of how calcium channel antagonists interact with the Cav1.2 channel, elucidating the nuances of their blocking activity and providing a clearer picture of their structural effects within the channel's pore domain.
January 2021
Journal of Graphics and Modelling
Discovery of dengue virus inhibitors with cutting-edge structure-based and advanced QM/MM affinity estimation
Dengue virus outbreaks pose a significant global challenge, with the virus continuing to spread due to rapid transportation and urbanization. Despite years of research, there are no effective small molecule antiviral treatments for dengue. While a vaccine has been developed, its efficacy remains uncertain, leaving an urgent need for alternative treatments.
This study employs advanced molecular modeling simulations to identify potential inhibitors of the dengue virus envelope protein. Specifically, the study investigates compounds that bind within the β-OctylGlucoside (β-OG) pocket, a conserved region of the protein. To accomplish this, the researchers use a combination of docking, molecular dynamics simulations, and binding free energy calculations.
The outcome of this research reveals the discovery of ten new compounds that exhibit a significant reduction in dengue virus production, as determined through cell-based virological assays. Additionally, the study provides an in-depth structural analysis of these compounds, with a focus on their electrostatic and lipophilic binding energy contributions.
One noteworthy aspect highlighted by the study is the influence of the desolvation penalty, which limits the activity of some of these compounds. Understanding this aspect is crucial in optimizing the effectiveness of potential antiviral compounds.
In conclusion, this research marks a significant step forward in the search for small molecule treatments against dengue virus. By utilizing advanced molecular modeling techniques, the study identifies promising compounds that can inhibit dengue virus production. The structural insights provided by the analysis contribute to the rational design of selective and potent inhibitors against dengue virus, addressing the pressing need for effective treatments in the fight against this infectious disease.
July 2021
Journal of Graphics and Modelling
Rational design of small-molecule immune checkpoints' inhibitors
In recent times, the inhibition of the PD-1/PD-L1 and CTLA-4 pathways has revolutionized cancer immunotherapy, ushering in a transformative era in the field. This breakthrough led to the selection of monoclonal antibodies (MABs) specifically targeting PD-1 as the 'drug of the year' in 2013. These antibodies have proven effective in restoring the functionality of exhausted T cells, enabling them to recognize and eliminate cancer cells. However, it's essential to acknowledge that while these MAB therapies have shown promise, they come with a set of notable drawbacks, including their exorbitant cost and the occurrence of severe side effects.
Our research team has been dedicated to developing an alternative approach using small molecule inhibitors for these critical pathways. Compared to the existing MAB-based therapies, our small molecules have the potential to offer several advantages. They may present a more cost-effective treatment option, facilitate easier administration, and provide enhanced control over the therapeutic process, all while effectively combating various types of cancer.
In this context, we present our ongoing efforts aimed at achieving this goal. We aim to provide an overview of our research progress and present data on the promising compounds we have designed to target both the PD-1/PD-L1 and CTLA-4 pathways.
Our research represents a shift towards the development of small molecules as potential therapeutics, offering a promising alternative to the current monoclonal antibody-based treatments. We recognize the need for more accessible and manageable cancer therapies that can provide effective and targeted treatments while minimizing the drawbacks associated with existing approaches. Our research endeavors seek to address these challenges and contribute to advancing the field of cancer immunotherapy.
July 2018
Cancer Research
Solving the PD-1 challenge
The utilization of atomistic computational modeling in drug discovery has yielded remarkable success by enabling the prediction of drug-receptor interactions and the optimization of drugs in terms of potency, selectivity, and safety. However, this approach encounters significant challenges when it comes to forecasting protein-protein interactions and designing regulators for such interactions. In this report, we present a rare instance where cutting-edge computer simulations, guided by experimental data, have achieved the accurate prediction of a complex protein-protein interaction, marking a notable breakthrough in the field.
Our study revisits a prior discovery involving the complex formed by the human PD-1 protein and its ligand PD-L1. We compare our previous findings with the recently published crystal structure of this same complex, conducting a detailed side-by-side comparison of our computational model with the experimentally determined crystal structure. The striking agreement between our model and the crystal structure underscores the accuracy of our protein-protein interaction prediction workflow.
This achievement holds significant implications for future endeavors in the realm of protein-protein interactions. It serves as a promising example of how advanced computational simulations, when guided by empirical evidence, can successfully predict complex interactions between proteins. This success story not only validates the robustness of our prediction methodology but also suggests its applicability to similar challenges and problems in the field.
In summary, our report highlights a remarkable accomplishment in the realm of protein-protein interaction prediction, demonstrating that state-of-the-art computational modeling, when informed by experimental data, can accurately forecast sophisticated interactions between proteins. This breakthrough offers valuable insights and opens up new possibilities for the rational design of regulators for complex protein-protein interactions, ultimately advancing the field of drug discovery and protein engineering.
October 2017
Journal of Molecular Modelling
Exploring the diverse conformations of PD-L1
In our recent investigation, our primary focus was directed toward exploring the PD-1 immune checkpoint pathway, with specific emphasis on the PD-L1 ligand. We embarked on a detailed examination of the conformational dynamics exhibited by PD-L1. To achieve this, we leveraged principal component analysis (PCA) applied to existing crystallographic structures in combination with both classical and accelerated molecular dynamics simulations.
Our study was particularly concerned with pinpointing the maximum structural displacements occurring within the PD-L1 protein. These displacements were meticulously scrutinized by analyzing various PD-L1 crystal structures and monitoring molecular dynamics trajectories. Consequently, we were able to attribute these displacements to specific regions within the protein that displayed a high degree of flexibility.
Furthermore, our investigation extended to the conformational preferences exhibited by PD-L1 regarding the binding of small molecules. Notably, we identified a crucial methionine residue located within the binding site, which emerged as a pivotal player in the process of drug binding.
To provide a comprehensive understanding, we also delved into the binding mechanisms of PD-L1 when it interacts with other binding partners. These interactions were thoroughly dissected and elucidated from a computational perspective.
The data and insights we have presented in this study are intended to contribute to the ongoing endeavors aimed at discovering and developing effective therapies targeting the PD-1 immune checkpoint pathway. By unraveling the conformational dynamics, binding preferences, and interaction mechanisms of PD-L1, we aim to provide valuable support to the broader scientific community striving to advance our understanding of immune checkpoint pathways and their potential as therapeutic targets.
September 2017
Biochemistry
Protein-Protein Docking: Are We There Yet?
Protein-protein docking algorithms represent robust computational tools with the capacity to meticulously dissect atomic-level interactions between proteins. In this chapter, we embark on a comprehensive exploration of the theoretical foundations underpinning diverse protein-protein docking algorithms, elucidating their merits, limitations, and highlighting significant case studies pertinent to each method. Our coverage encompasses a wide array of methods characterized by distinct search strategies and scoring techniques, offering a holistic perspective on this vital field.
Among the methods to be discussed are exhaustive global search, fast Fourier transform-based search, spherical Fourier transform-driven search, direct exploration in Cartesian space, local shape feature matching, geometric hashing, genetic algorithms, randomized search approaches, and Monte Carlo-based exploration. Each method brings its unique strengths and considerations, and we will delve into their specific attributes in detail.
Furthermore, we delve into the intriguing realm of protein flexibility and how it can be seamlessly integrated into the docking procedure. The ability to account for protein flexibility is pivotal in achieving more accurate docking results, and we will explore the various strategies employed to incorporate this vital aspect.
As we navigate through the intricacies of protein-protein docking, we also turn our gaze toward the future. In this chapter, we contemplate potential directions for further advancement in the field, envisioning ways to enhance and refine the existing methods. These forward-looking insights will encompass innovative strategies and approaches that promise to elevate the precision and efficacy of protein-protein docking algorithms.
Ultimately, our aim in this chapter is to provide a comprehensive and insightful overview of protein-protein docking algorithms, encompassing their foundational concepts, practical implementations, and future prospects. By doing so, we hope to contribute to the evolving landscape of protein-protein interaction analysis and its profound impact on various fields, including drug discovery and structural biology.
December 2016
Methods and Algorithms for Molecular Docking-Based Drug Design and Discovery
Towards Modelling The Full human Programmed Death-1 (PD-1) Pathway
mmune checkpoints represent pivotal regulators of the immune system, responsible for maintaining a delicate balance between preventing excessive immune responses and combating chronic infections and cancer. The groundbreaking strategy of inhibiting these immune inhibitory checkpoint pathways has recently emerged as a transformative approach in the realms of cancer and antiviral immunotherapy. To unlock the full potential of this approach, it becomes essential to model these pathways at the atomic level, offering a foundation for the rational design of selective blockers.
Currently, crystal structures available for immune checkpoints are predominantly non-human, presenting a significant limitation in understanding their interactions. Our research team has undertaken the crucial task of constructing atomistic models for these immune checkpoint proteins. Our goal encompasses characterizing their intricate protein-protein interactions and innovating new inhibitory drugs to modulate their activities effectively.
This article spotlights our recent investigation, where we concentrated on modeling the human Programmed Death-1 (hPD-1) pathway. We meticulously examined the interactions between hPD-1 and its two human ligands, shedding light on their distinct binding patterns. Notably, we discerned that hPD-1 binds differently to each of its ligands, revealing intricate nuances in the mode of interaction. Furthermore, our study unveiled that the binding modes of hPD-1 to its ligands in humans contrast with those observed in mouse models, underscoring the limitations of current crystal structures primarily derived from mice.
Our findings represent a substantial leap forward in comprehending the intricate receptor-ligand interactions within the PD1 pathway. They serve as a critical milestone toward constructing a comprehensive model for the entire PD1 pathway, a pursuit poised to significantly bolster ongoing endeavors aimed at developing antibodies and small molecule drugs targeting this pivotal T cell immune-regulatory mechanism.
In essence, our research contributes to unraveling the complexities of immune checkpoints, providing valuable insights that hold the promise of advancing immunotherapy and ultimately improving treatment outcomes for cancer and viral infections.
November 2015
Receptors & Clinical Investigations
Entropy in bimolecular simulations
Entropy of binding plays a significant role in biomolecular interactions, often exerting a substantial influence on binding affinity. In many instances, it can be a challenging and even detrimental component of these interactions. While calculating the enthalpic part of binding free energy is relatively straightforward, estimating binding entropy proves to be considerably more complex. To achieve a precise evaluation of entropy, one would ideally need to comprehensively explore the entire phase space of the interacting biomolecules. However, this task remains exceptionally difficult within the confines of conventional molecular simulations, necessitating various approximations in entropy calculations.
Numerous established methods have been developed to estimate binding entropy, with a primary focus on reliably assessing the conformational component of entropy. In this review, we delve into these methods, placing particular emphasis on techniques that leverage atomic fluctuations to extract entropy. We meticulously elucidate the theoretical underpinnings of each method, shedding light on their respective strengths and limitations. To provide a comprehensive understanding, we also present a range of case studies for each of these entropy estimation methods.
Our aim in presenting this concise yet all-encompassing review is to offer a valuable resource for comprehending these entropy estimation methods. By delving into the theoretical foundations and practical considerations of each approach, we aspire to equip readers with the knowledge necessary to navigate the complexities of entropy calculations in biomolecular interactions. This review serves as an informative tool for researchers and practitioners in the field, enhancing their ability to address practical challenges that may arise during such calculations.
In summary, our review endeavors to elucidate the intricacies of estimating binding entropy, a critical aspect of biomolecular interactions. By exploring various methods and their associated case studies, we aim to facilitate a deeper understanding of the theoretical frameworks, applications, and limitations inherent in these entropy calculations.
November 2015
Receptors & Clinical Investigaions
Identifying how human PD-1 uniquely binds to its ligands
Immune checkpoints serve as crucial regulators within the immune system, diligently maintaining the delicate balance between preventing persistent immune activation and safeguarding against chronic infections and cancer. Recently, the strategy of obstructing immune inhibitory checkpoint pathways has emerged as a transformative approach in both cancer and antiviral immunotherapy, marking a significant paradigm shift. To unlock the full potential of this approach, it becomes imperative to delve into the atomic-level modeling of these pathways, paving the way for the rational design of precise and selective blockers.
Unfortunately, the current landscape of crystal structures available for immune checkpoints primarily encompasses non-human sources and remains restricted in the depth of interactions they capture. In response to this limitation, our research team has dedicated its efforts to constructing atomistic models of these critical immune checkpoint proteins. Our objectives revolve around elucidating their intricate protein-protein interactions and pioneering the development of novel inhibitory drugs tailored to modulate their activities effectively.
This article serves as a spotlight on our recent study, where we honed in on modeling the human Programmed Death-1 (hPD-1) pathway. Our investigation meticulously characterized the interactions between hPD-1 and its two human ligands, yielding invaluable insights. Notably, we uncovered that hPD-1 exhibits distinct binding modes when engaging with each of its ligands. Additionally, our study underscored a critical distinction between the binding modes observed in mouse and human models, accentuating the limitations of relying solely on mouse-derived crystal structures.
Our findings represent a pivotal leap forward in unraveling the complexities of receptor-ligand interactions within the PD1 pathway. They serve as a critical milestone in the endeavor to construct a comprehensive model that encapsulates the entirety of the PD1 pathway. This advancement undeniably fuels the ongoing efforts to develop antibodies and small molecule drugs, opening up new avenues for targeting this pivotal T cell immune-regulatory mechanism.
In essence, our research contributes to expanding our understanding of immune checkpoints and their potential as therapeutic targets. By delving into the intricacies of receptor-ligand interactions and binding modes, we empower the scientific community to forge ahead in the quest to harness the potential of immunotherapy for the benefit of patients worldwide.
April 2015
Journal of Molecular Graphics and Modelling