Qubit Pharmaceuticals accelerates and transforms drug discovery by publishing an unprecedentedly accurate computational approach combining their core technologies with exciting innovations: highest accuracy in molecular simulation thanks to next generation Polarizable Models and Tailored Free Energy Methods to predict the binding affinities between RNA and small molecules, the newly developed Lambda-ABF scheme and enhanced sampling techniques, including a Machine Learning approach. Our new pipeline overcomes the major challenges faced by current tools in the specific context of RNA and represents a promising yet relatively unexplored avenue for the design of new drugs.
Before delving into the details, it’s crucial to grasp the significance of RNA targets in advancing drug discovery. These targets represent an innovative frontier, allowing for the development of drugs that can affect proteins typically deemed difficult to reach, thereby broadening treatment avenues for complex diseases like cancer, neurodegenerative, inflammatory and viral diseases. Our paper introduces a pioneering computational approach that accurately predicts the binding of small molecules to RNA, effectively addressing the challenges of simulating RNA dynamics. By leveraging advanced Polarizable Models, Tailored Free Energy Methods, and our newly developed Lambda-Adaptive Biasing Force (Lambda-ABF) approach, we can precisely identify promising drug candidates. This expertise not only enhances our understanding but also positions us at the forefront of RNA-targeting therapeutics, inviting forward-thinking partnerships that can drive meaningful advancements in the field.
We would like to express our sincere gratitude to our partners, Bpifrance, European Innovation Council, FRANCE 2030, GENCI, Secrétariat général pour l'Investissement, Région Île-de-France, and Sorbonne Université, whose support has been pivotal in enabling our research and pushing the boundaries of computational drug discovery into exploring RNA as therapeutical target.
RNAs are of prime importance in many biological and pathological processes. For instance, RNAs are involved in several infectious diseases (Hepatitis C, Dengue…) (ref), genetic diseases (Myotonic Dystrophy Type 1, Fragile X Syndrome…) (ref) as well as in cancers (ref). Thus, RNA-targeting therapeutics offer an attractive alternative to reach traditionally undruggable proteins and expand the druggable target space in the human genome from 0.2% to ~70%[1].
Among strategies developed to target RNAs, the “small-molecule’’ approach is particularly promising as it effectively reaches RNA targets while avoiding major adverse reactions such as off-target effects, interaction with immune system… Moreover, small-molecule therapeutics are more likely to be transposed into drugs with potential for oral bioavailability and penetration of blood–brain barrier. That said, to develop therapeutical new “small-molecules”, it is essential to run accurate preliminary computational simulations to understand the structural dynamics of the system as well as its mechanism of action and interplay with other partners. However, RNA simulation is particularly challenging. Indeed, the structure of RNA is notoriously complex to study, as it is highly negatively charged, flexible and quick to adopt different 3D conformations (different three-dimensional shapes) depending on its environment and interactions. What's more, most of the time, the sites of interaction between RNA and small molecules are on the surface, in direct contact with the RNA's immediate environment (ions, polarizable water molecules, etc.). These sites are therefore particularly sensitive to variations in the context in which the interaction between RNA and small molecules occurs.
Altogether, these parameters complicate precise targeting and simulation of RNA in the absence of a computational model that deals accurately with electrostatics.
In short, targeting RNA matters because it can help create new drugs for diseases that are hard to treat, but, like anything that is worth doing, it is incredibly complex and requires highly advanced technology.
We propose a new tailored computational protocol that overcomes the technical limitations surrounding the simulation of RNA in complex with small molecules (meaning they interact with each other or associate together as a stable structure). Our protocol can efficiently predict the likelihood, effectiveness, and modes in which a small molecule interacts with its RNA target.
First, it was essential to resort to a computational model that encompasses an accurate treatment of electrostatics, allowing to reproduce, and with high accuracy, the interactions of the negatively charged RNA with its environment (polarizable water molecules, ions and small molecules). This is the case of AMOEBA, a second-generation force field parameterized using Quantum Chemistry methods, and which accounts for many-body polarization effects.
The AMOEBA computational model allows us to accurately reproduce the interaction of RNA and small molecules with their environment, which is essential because it captures the complex electrostatic interactions and polarization effects that influence how these molecules bind together. Highest levels of accuracy are crucial for identifying the binding modes of small molecules and predicting their binding affinity to the RNA target, enabling more effective drug design.
To identify the possible 3D conformations of RNA and small molecules, and their propensity to bind together, we combined the advanced AMOEBA polarizable force field, to the newly developed Lambda-ABF scheme associated to refined restraints allowing for efficient sampling. These molecular dynamics (MD) simulations are done with the GPU-accelerated Tinker-HP software. To ensure consistency between our simulations and the biological systems they mimic, and thus limit the production of artifact results, we applied DBC collective variables (Distance-to-Bound-Configuration)[2], allowing us to apply positional, orientational, and conformational restraints to sampled small molecule structures.
This synergy not only overcomes the limitations of available free energy methods but also establishes a new standard in computational modeling of RNA-ligand interactions with unparalleled accuracy and speed.
Our protocol enhances the prediction of 3D binding conformations for RNA and small molecules, including even the rarest ones previously unreachable. This unprecedented access significantly improves the relevance of binding predictions. Additionally, the newly developed Lambda-ABF method for alchemical free energy simulations delivers robust binding affinity predictions that align with experimental results. Stronger RNA-small molecule binding increases the likelihood that the small molecule will effectively impact its RNA target.
The structures of bound (HOLO) and unbound (APO) RNA show that the RNA undergoes a dramatic ligand-induced conformational adaptation. This APO-HOLO conformational change involves complex collective motions that are challenging to capture with standard methods. To determine the free energy barrier that RNA must overcome to adopt these conformational changes, we applied machine learning to identify effective collective variables (Deep-LDA) for use in enhanced sampling simulations based on an evolution of metadynamics (OPES-explore).
This approach complements our strategy by providing insight into the free energy barrier associated with the APO-HOLO conformation change for additional validation. By employing machine learning-based collective variables, in combination with an evolution of metadynamics, we can further enhance sampling to capture the challenging RNA conformational changes. This means we can better understand how RNA changes shape when interacting with small molecules, which is crucial for developing effective RNA-targeted drugs.
We tested our strategy on one of the most challenging RNA-small molecule system for molecular dynamics simulations in the context of binding mode identification and binding affinity prediction: the viral RNA sequence involved in the Hepatitis C pathogenesis (HCV-IRES IIa) and the currently available drugs acting as inhibitors (2-aminobenzimidazole derivatives). This RNA-small molecule was chosen as study case because of its complexity that makes it one of the most challenging systems for atomistic simulations and absolute binding affinity calculations. Our results correlate with traditional experimental data, confirming the robustness and relevance of our approach.
By successfully applying our approach to the complex Hepatitis C RNA-small molecule system, we demonstrate that our method can accurately predict how well potential drugs bind, which is crucial for developing effective treatments for difficult-to-treat diseases.
In conclusion, the present work represents a new achievement in computational methodologies for drug design, specifically targeting the complexities of RNA. By integrating advanced technologies, developed by Qubit Pharmaceuticals and their partners, like the Lambda-ABF method, and Tinker-HP Software, powered by Qubit Pharmaceuticals’ 200 GPU supercomputing cluster, Gaia, and run on its proprietary drug discovery platform Atlas, we achieve unparalleled accuracy and speed in predicting RNA-small molecule interactions. We also included machine learning techniques to further enhance sampling and capture the challenging RNA conformational changes. This innovative approach not only enhances our understanding of RNA-targeting mechanisms but also lays a robust foundation for developing RNA-targeted therapeutics. As pharmaceutical companies seek to develop effective treatments for diseases that currently lack viable options, our research provides the insights and tools necessary for precise drug discovery, ultimately guiding the way toward new breakthroughs in healthcare.
Massive Congratulations to the authors of the paper this blog is based on:
Narjes Ansari ,Chengwen Liu ,Florent Hedin , Jérôme Hénin , Jay W. Ponder , Pengyu Ren , Jean-Philip Piquemal ,Louis Lagardère , Krystel El Hage
Read the preprint of this paper here: https://chemrxiv.org/engage/chemrxiv/article-details/66e6b6bb12ff75c3a16485ed
[1] Source: https://pmc.ncbi.nlm.nih.gov/articles/PMC6420209/
[2] DBC stands for Distance-to-Bound-Configuration and refers to a specific set of restraints used on a group of atoms of the ligand in alchemical free energy simulations for efficient sampling.