The following posters have been proposed for the upcoming UKQSAR Spring Meeting on 15th April 2025:
| Name | Organization | Poster Title | Poster Abstract |
| Zexiang Han | University of Cambridge | High-throughput formulation screening for therapeutic antibodies. | Therapeutic monoclonal antibodies (mAbs) are essential for treating diseases such as cancer and autoimmune disorders, but ensuring their stability and solubility at high concentrations remains a significant challenge. Excipients additives that stabilize these drugs are crucial for preventing aggregation, yet identifying the optimal excipients for a given mAb often relies on trial and error. Striking a balance between maintaining the antibody’s binding specificity and minimizing self-association is vital to preserving therapeutic efficacy. In this study, we leveraged combinatorial droplet microfluidics to systematically investigate the effects of common pharmaceutical excipients on the solubility of clinical IgG antibodies. Using high-resolution solubility phase diagrams, each containing tens of thousands of data points, we quantified excipient-specific effects both within individual antibodies and across multiple mAbs. Our findings revealed that excipient effects are highly specific and not broadly applicable across different antibodies. We leveraged sequence analysis and structural prediction tools to identify key physicochemical traits influencing excipient response. This offers a potential route to tailored formulation development and reducing the reliance on extensive empirical testing. These insights into excipient-protein interactions provide a foundation for more rational biotherapeutic formulation design, paving the way for improved stability and efficacy of therapeutic antibodies and other protein-based drug modalities. |
| Andrew Henry | Chemical Computing Group | Constrained Peptide Modeling, Conformational Analysis and Property Predictions. | Peptides play an integral role in a myriad of biological pathways. Given their geometric and chemical diversity, peptides can effectively bind a broad spectrum of biological targets. This inherent diversity also presents numerous challenges such as the size of amino acid sequence space, assessment of low energy peptide conformational states and prediction of peptide properties. In this work, we present in silico methods for building peptide models, sampling peptide conformations and predicting the properties of peptides. |
| Shayantan Chaudhuri | University of Nottingham | Calculating Covalent Inhibition Using Quantum Chemistry, Hybrid QM/MM and Quantum Computing | Abstract: Covalent drug design presents a challenging frontier for conventional computational resources. The size of the system typically needed to study covalent inhibitor ligand-protein binding limits the applicability of high-level ab initio methods, and computational chemists must make concessions via various approximations. This bottleneck limits progress in the burgeoning field of covalent drug design. We have used high-level quantum chemistry calculations to characterise the nucleophilic addition of an archetypal nucleophile, methanethiolate, to various nitrogen-containing Michael acceptors which are representative of some small-molecule covalent inhibitors of interest. We investigated the structural, energetic, and electronic properties of the resulting enolates, as well as their reaction profiles in gas phase and within continuum solvent models. In order to assess the influence of the receptor protein on the binding thermodynamics, we are studying related covalent inhibitors in complex with the protein CDK12, using hybrid QM/MM calculations through the QM/MM features of NAMD interfaced to ORCA as the QM package. The most forward-looking part of our work focuses on the opportunity to use quantum computing to model molecular interactions for drug discovery. We are currently working in collaboration with the quantum algorithm company, Phasecraft, and the quantum computing company, QuEra on a novel pathway to quantum-accelerated computations of covalent inhibitors on near term quantum devices by extending advances in quantum algorithm design of low depth and compact quantum circuits to molecular systems. |
| Terence Egbelo | University of Sheffield | Improving Target-Adverse Event Association Prediction by Mitigating Topological Imbalance in Knowledge Graphs | The work tackles the prediction of new target-AE associations as KG completion using a large-scale biomedical knowledge graph. It also showcases a novel approach to handle topological bias in KGs. Our procedure was found to produce significantly better predictive performance on the most sparsely connected targets (accuracy improvement of ~0.15 on the bottom 15% targets by no. of AE associations) than an existing conventional alternative, the Degree-Weighted Path Count, or DWPC, first reported by Himmelstein et al (2015) and thereafter used e.g. by Himmelstein et al (2017) and Binder et al (2022). We then used the metapath-based KG completion approach, as improved by the new modification, to demonstrate prediction interpretability, including in cases where the unmodified and DWPC alternatives make errors. |
| Rob Ziolek | Kvantify | Predicting and understanding drug-target binding kinetics via molecular simulations | Drug-target residence time is commonly put forward as a key indicator of drug efficacy and a valuable metric for assessing features such as selectivity, toxicity, and dosage. Nevertheless, computational tools for predicting residence times have still not been widely adopted into real-world computer-aided drug design projects. While a variety of accurate and efficient molecular dynamics (MD) based approaches are under active development, their applicability within a high-throughput setting is still limited by the large differences in time scales between (un)binding events and integration step size. We have developed koffee: a fully physics-based tool for scoring ligands or protein variations according to residence time. koffee requires no training data or target-specific calibration and reduces the required number of function evaluations by 2-3 orders of magnitude compared to corresponding MD-based approaches. The accuracy of the tool has been demonstrated on a variety of drug targets and use cases including ranking of ligands for HSP90 and prediction of the effect of binding-pocket mutations on the M3/tiotropium complex. |
| Hannah Turney | King’s College London | Software toolkits for in silico screening of polymer excipients used in small molecule formulation and drug delivery. | The use of polymers as excipients in small molecule pharmaceutical formulations is an established approach for the controlled delivery of drugs. However, designing safe and effective formulations is resource-intensive and delays product delivery to the clinic, primarily due to the sensitivity of polymer substructure to delivery properties. / Molecular dynamics (MD) simulations are a powerful in silico tool that can provide detailed insights into the time-dependent behavior of molecular systems at the atomic level. MD can reveal information about the conformational changes, binding interactions, and dynamical properties of molecules. Although MD simulations are associated with high computational demand, increased access to computational power has allowed for the construction of longer and larger simulations. However, limitations remain in the routine application of MD simulations to polymer excipients, such as the difficulty in parameterizing larger polymers and ensuring the transferability of force fields across different polymer chemistries. In this project, we create a robust and scalable building and parameterization workflow for polymer excipients. This research uses advanced molecular dynamics techniques from organizations such as the Open Force Field Consortium to create a robust in silico polymer parameterization methodology. With our established workflow, we perform systematic molecular dynamics simulations of different polymer:drug systems to yield kinetic and mechanical parameters predicting excipient suitability for drug delivery. This enables fast, accurate, and reproducible profiling of polymers in the context of drug formulation design. We collaborate with experimental formulation development teams at Johnson&Johnson Innovative Medicine to validate our models and drive the design of our polymer:drug candidate systems. By incorporating existing open-source software, and sharing any resulting tools developed, we aim to promote reproducibility and collaboration throughout the project. |
| Max Winokan | Diamond Light Source | A Formulaic Approach to Fragment Progression. | Hit Interaction Profiling for Procurement Optimisation (HIPPO) is a multi-objective optimisation tool for selecting effective experiments in structure-based drug design. Structures from high-throughput fragment screens present a rich starting point for derivative compounds that aim to recapitulate observed interactions[1] . Even at lower-molecular weights, the combinatorial possibilities rapidly explode when exploring expansion vectors in and beyond REAL space (>1010 compounds), where single step chemistry can unlock millions of synthetic routes from a dozen fragment merges[2] . Despite the advancements in robotic automation, the capacity for rapid screening remains on the order of thousands of compounds, thus effective selection of compounds remains a problem. Manual review of algorithmic merges and expansions can lead to higher in-crystal hit rates but is impractical at scale. Our approach hinges on the premise that merges and expansions of fragment screen hits are by nature very weakly bound and hence per-compound scoring algorithms can miss exploration opportunities[1,3] . HIPPO provides generation, scoring, and optimisation of synthetic routes via customisable parameters that balance recapitulation of experimental interactions, SAR exploration, and procurement logistics. By scoring sets of virtual hits rather than ranking them individually, our approach reduces the risk of over reliance on docking, provides efficient hit validation and balances multiple hypotheses to increase the information gain. HIPPO empowers users to create a bespoke approach for their drug discovery campaign and presents an accessible HTML output that rationalises the selection with interactive graphs that compare the risk, opportunity, and value-for-money of different optimised proposals. Integration into the Fragalysis platform will enable its users to optimise follow-up chemistry in real-time. A trial campaign with a calcium binding the protein target (Frequenin) has demonstrated that random samples of synthetic routes are inefficient, as it resulted in low numbers of product compounds that fail to leverage the interactions seen in crystallographic. We demonstrated that certain base compounds provided more cost-effective routes to SAR exploration. Our pipeline was able to identify a sample of synthetic routes that leveraged inexpensive building blocks (<$7/mg), produced 60% greater coverage of features in the experimental ensemble and quintupled new interactions when compared to random samples within the same $8000 budget. [1] Keserq, G. M., & Makara, G. M. (2006). Hit discovery and hit-to-lead approaches. Drug discovery today, 11(15-16), 741-748. [2] Sadybekov, A. A., Sadybekov, A. V., Liu, Y., Iliopoulos-Tsoutsouvas, C., Huang, X. P., Pickett, J., … & Katritch, V. (2022). Synthon-based ligand discovery in virtual libraries of over 11 billion compounds. Nature, 601(7893), 452-459. [3] Reynolds, C. H., & Holloway, M. K. (2011). Thermodynamics of ligand binding and efficiency. ACS medicinal chemistry letters, 2(6), 433-437 |
| Ruslan Kotlyarov | University of Cambridge | NMR Prediction Uncertainty Enables DFT-Free Structure Confirmation | |
| Daniella Hares | Institute of Cancer Research | Enhancing Small Molecule Binding through Computational Analysis of Water Networks | Water molecules in the binding site can have a critical role in small molecule binding to proteins and are an important consideration in structure-based drug design. Water networks have additional complexity as displacing one water molecule has subsequent effects on the remaining network. When modifying the structure of a lead compound, the disruption of a water network can have beneficial or detrimental effect on potency and this outcome is impossible to determine experimentally without time-consuming synthesis of the new compound. Computational methods can be used to study these water networks and provide insights that are able to guide forward synthesis. We used Grand Canonical Monte Carlo (GCMC) to prospectively design BRD9 inhibitors that optimally perturbs a water network. Building on a scaffold from the literature, we used a generative AI model to design substituents that pointed towards a water network in the KAc binding site. Displacing or rearranging this water network has been used by other groups to gain selectivity for a subset of bromodomains, such as SMARCA4 and TAF1(2), but is unprecedented for BRD9. The effect of the designed compounds on the binding free energy of the water network was predicted by GCMC and used to shortlist compounds for synthesis. By showcasing this application, we hope to encourage the use of GCMC more widely in drug design when working with water networks in the binding site. |
| Long-Hung Pham * | Imperial College London | Preferential Multi-Objective Bayesian Optimization for Drug Discovery | |
| Mike Bodkin | Dundee University | TBA | |
| Peter Ibrahim | Drug Discovery Unit, University of Dundee | FraGrow From Fragments to LLMs | |
| Lukas Eberlein | OpenEye / Cadence | Machine learning enhanced small molecule conformational sampling: Thompson sampling in OMEGA | |
| Greg Price | BIOVIA | Bayesian Optimization in BIOVIA Pipeline Pilot | |
| Jiajun Zhou | Imperial College London | Deep generative design of porous organic cages via a variational autoencoder | |
| Leonie Windeln | University of Southampton | Turning toxins into targets: design of alpha±-conotoxin capturing proteins alpha±- | Conotoxins are naturally occurring peptides, that are of pharmacological interest, but their potency also poses risks without available anti-toxins. Inspired by recent advances in AI-driven protein design, we generated proteins that bind to alpha±-Conotoxins. For this we tested different scaffold constrains and applied stringent filtering for relevant protein-peptide interactions. Early experimental validation studies of our best designs have been successful, and we now work on improv these using AI as well medicinal chemistry methods. |
| Martijn Bemelmans | J&J Innovative Medicine | University of Geneva | Hunting for cryptic pockets through enhanced sampling of protein dynamics. | Despite decades of research, some proteins remain pharmaceutically intractable, including phosphatases, transcription factors, and GTPases like K-Ras. To engage these “undruggable” targets with small molecules, the concept of allosteric modulation via cryptic pockets provides an innovative solution. However, due to the short lifespan of conformations with such sites exposed, their location or even existence is hard to find both experimentally and computationally. The emerging method in molecular simulations called “on-the-fly probability enhanced sampling” (OPES) allows quick sampling of protein conformational states that can then be subjected to pocket prediction algorithms to find binding sites in newly uncovered conformational states. Developing compounds to bind these new binding sites then has the potential to “lock” the protein in this uncovered state, providing an opportunity for (allosteric) modulation. |
| Tim Hindges | MSD | Benchmarking the Fragment Molecular Orbital (FMO) method for Tyrosine Kinase 2 (TYK2) and Cyclin-Dependent Kinase 2 (CDK2) datasets | |
| MarcusGastreich and Filomena Perri | BioSolveIT GmbH | Chemical Space Docking! Mining Billion-Sized On-Demand Collections for ROCK1 and PKA Inhibitors | Chemical space docking is a novel approach in the toolbox of virtual screening: It can handle massive numbers of molecules from on-demand chemical spaces such as the Enamine REAL Space. In analogy to known virtual screening methods |
| Thomas Potter | Cresset | Absolute binding free energy calculations for accurate activity prediction in modern scaffold hopping experiments. | Accurate prediction of binding affinity is an important part of modern virtual screening (scaffold hopping) workflows, though use of the popular relative binding free energy (RBFE) approach is limited by its reliance on high chemical similarity between molecules. Absolute binding free energy (ABFE) allows for calculation of accurate binding affinities without the requirement for chemical congenericity, though they are typically difficult to set up and expensive to run. Modern, single platform modelling suites increase accessibility to users by removing technical barriers, as well as offering tools that reduce computational cost whilst retaining computational rigour. We present here a prototypical workflow for scaffold hopping and triaging in Cresset s Flare! molecular modelling suite. We make use of tools such as Grand Canonical Nonequilibrium Candidate Monte Carlo (GCNCMC) sampling for water molecule placement and Adaptive Lambda Scheduling (ALS) for bespoke on-the-fly lambda coordinate scheduling |