Newsletter

Archived Newsletters

UK-QSAR Autumn 2025

The UK-QSAR and Cheminformatics Group

Welcome to the Autumn 2025 UK-QSAR Newsletter!

 

The Autumn UKQSAR and Cheminformatics Society Meeting will be held at the AstraZeneca Discovery Centre, Cambridge on Thursday the 13th of November. Registration has now closed but a waiting list is in operation so please continue to register your interest in attending via the registration link here, and if you are already registered but can no longer attend, please let us know so we can free up your slot. The meeting is organised and hosted by AstraZeneca and the theme of this meeting is “Molecules in Motion”, including talks on physics-based modelling, AI-enhanced structure-prediction and more.

Please note that paper copies of the agenda and abstracts will not be provided onsite, so we recommend that you bookmark this page on your device or screenshot the agenda section below.

Our newsletter contains more details of the meeting venue below together with the draft agenda,  travel and accommodation information, info for  poster presenters and  talk abstracts/pre-reading material, as well as the usual sections on  job openings and upcoming  meetings which may be of interest. 

Our Spring Meeting 2026  will be held in Edinburgh on 21st April 2026, and will be jointly hosted by Enamine and BioSolveIT. Save the date and more details will be posted nearer to the event.

Autumn Meeting Information

REGISTRATION

Registration is now closed, but you can be added to the waiting list by registering your interest here.

AGENDA

UKQSAR, Autumn Meeting, AstraZeneca, Cambridge, 13th November 2025 Program

09:00- 10:00 Open registration, coffee/tea
10:00 -10:10Welcome and opening remarks (Graeme Robb)
Session 1 (Chair: Hannah Bruce MacDonald)
10:10-10:40Grand Canonical Simulations for In silico Prediction of Fragment Binding Sites, Modes, and Affinities
Jon Essex – University of Southampton
10:40-11:10Beyond Affinity: Leveraging MD Simulations to Compute Drug-Target Unbinding Kinetics
Christina Athanasiou – AstraZeneca
11:10-11:40Multiscale Simulations of DNA – How Atomistic MD Simulations Inform Systems Biology Models of Bacterial Gene Circuits and Their Regulation
Sarah Harris – University of Sheffield
11:40-12:10Computational Approaches for Targeted Protein Degradation in Cancer Drug Discovery
Andrea Scarpino – ICR
12:10-13:30Lunch and poster session
Session 2 (Chair: Silvia Bonomo)
13:30-14:00Federated Learning for Protein Co-Folding Models Using Open-Source AlphaFold3 Derivatives
Benedict Irwin – Apheris
14:00-14:30Open Science for Molecular Modelling with the Open Force Field Initiative
Daniel Cole – Newcastle University
14:30-15:00Residence-time Scoring for High-Throughput Computer-Aided Drug Design
Daniela Dolciami and Rob Ziolek – Kvantify
15:00-15:30Coffee break and poster session
Session 3 (Chair: Ilenia Giangreco)
15:30-16:00Active Learning FEP for Prospective Use in Drug Design
Jack Glancy – GSK
16:00-16:30Modern Hit-Finding with Structure-Aware De Novo Design: Identification of Novel A2A Receptor Ligands Using Reinforcement Learning
Pierre Matricon – Nxera Pharma
16:30-16:40Concluding remarks, poster prize winner announcement and close (Nathan Brown)

PRE-MEETING MEAL

There is a plan for folks local to Cambridge or who are arriving the night before the meeting to meet in The Station Tavern from 6:30pm. Numbers for the meal have already been confirmed but please contact Silvia Bonomo if you can no longer make it or if you forgot to register for the meal.

ABOUT THE AZ Discovery Centre (The DISC)

The bold, modern design of the DISC has earned its architects, Herzog & de Meuron / BDP, both a national and a regional RIBA award in 2025, and the building was shortlisted for this year’s Stirling Prize.

An obstructed view of a modern office courtyard with a tree with yellow leaves

The building sits within the Cambridge Biomedical Campus, close to the Addenbrookes Cambridge University teaching hospital and Royal Papworth, the MRC and CRUK’s Cambridge Research Institute, and houses more than 2000 researchers.  The building was designed for energy efficiency, and in its first full year of operation its carbon footprint was similar to just three average UK households.

LOCATION AND TRAVEL 

For those planning their travel, we encourage the use of public transport and other green travel options. If arriving by train at Cambridge rail station, you have several options to reach The Discovery Centre:

  • By bus: The Bus A busway from the station runs every 15 minutes and takes about 10 minutes to reach The Discovery Centre.  Allow extra time during commuting hours.
  • By bike: Cycling from the rail station to The Discovery Centre takes around 10 minutes.  Visitor cycle racks are situated outside the building on Robinson Way.
  • On foot: It takes approximately 40 minutes to walk from the rail station to The Discovery Centre via Hills Road.

If you plan to arrive by car, there are a few options to choose from.

  • There is a multistorey carpark near The Discovery Centre. Please look for ‘Car Park 2’ on Robinson Way. This will cost £20-26 for all-day parking.
  • The Trumpington Park and Ride is a free option.  This is a 5 minute bus ride (every 10 minutes), 10 minute cycle or 30 minute walk from the venue.
  • The Babraham Road Park and Ride is another free option.  This is also a 10 minutes cycle or 30 minute walk to the venue. There is no direct bus connection.

 

ACCOMMODATION

There are a number of hotels situated close to the rail station and therefore a short walk from the Discovery Centre. The first three options below are larger hotels and likely to have room, but there are plenty of other options. The organising committee haven’t stayed at any of these hotels, so please don’t take this as a recommendation of quality.

This is further from the station but still close to the venue. May be suitable if arriving by car.

  • Lord Byron Inn [££], 22 Church Lane Trumpington, Cambridge, CB2 9LA

And finally, the University Rooms offer a different accommodation experience – some will be closer to the venue than others:

FOR POSTER PRESENTERS

Abstract submission is now closed.

Please note that poster-boards are portrait-format. Each board is 880mm wide and 1770mm tall. This is large enough for a portrait A0 poster or a landscape A1 poster. 



Abstracts & Pre-Reading Material

Some speakers have provided the following abstracts and references which might be of interest ahead of their talks.

Jon Essex, University of Southampton

Grand Canonical Simulations for In silico Prediction of Fragment Binding Sites, Modes, and Affinities

Fragment based drug discovery (FBDD) is widely used in the pharmaceutical industry as a route to generating lead compounds. Through knowledge of their binding sites, fragment hits may be combined into single molecules with good potency and physical properties. Computational FBDD supports experiment by providing a route to library design, virtual screening, binding site identification and binding affinity prediction.

In this talk, the development and application of grand canonical non-equilibrium Monte Carlo (GCNCMC) in this domain will be discussed. By combining non-equilibrium move proposals, with grand canonical Monte Carlo acceptance tests, we are able to insert and delete small molecules into the binding sites of host-guest and protein systems, much more efficiently than conventional molecular dynamics. Through these simulations we are able to identify potential ligand binding sites by augmenting the sampling in mixed-solvent molecular dynamics. Fragment binding poses, including situations where multiple binding poses have been reported, are also readily identified. Finally, by varying the chemical potential of the simulations, absolute ligand binding free energies may be calculated, without the need for restraints or corrections to address multiple binding poses. Binding sites, poses, and affinities may all be calculated through a single series of simulations run at different chemical potentials.

Christina Athanasiou, AstraZeneca

Beyond Affinity: Leveraging MD Simulations to Compute Drug-Target Unbinding Kinetics

The study of kinetics, i.e. the rates at which drugs associate with and dissociate from their biological targets, plays a critical role within drug discovery and pharmacology. Although binding affinity has traditionally received primary consideration, a growing body of research underscores the significance of binding kinetics, which may exert equal or greater influence on a compound’s efficacy, safety, and overall therapeutic profile. In certain therapeutic contexts, such as tissue targeting, candidate selection often prioritises those exhibiting the slowest dissociation rates. However, reliable—yet computationally efficient—prediction of drug–target residence time, the inverse of the dissociation rate, remains a considerable challenge and is consequently seldom incorporated during the early stages of drug design. In this work, we benchmark two distinct implementations of the Random Acceleration Molecular Dynamics (RAMD) enhanced sampling approach for their ability to predict residence times in the context of live drug discovery campaigns.

Sarah Harris, University of Sheffield

Multiscale Simulations of DNA – How Atomistic MD Simulations Inform Systems Biology Models of Bacterial Gene Circuits and their Regulation

Biomolecular simulations are used both to understand biological mechanisms, and to predict the outcome of biological interactions or processes, assuming that they can be appropriately validated. When applied to DNA topology, computer simulations face the challenge that while supercoiling affects DNA structure at the level of the twist of DNA base pairs, the consequences are amplified up to the genomic level. Even with the most powerful supercomputers, it is not possible to simulate the whole nucleus in atomistic detail over the timescales needed to observe the transcription of a gene. Therefore, it is necessary to use multiscale modelling that integrates representations at different resolutions. Here I will show how we have used atomistic simulations to inform the construction of abstract models of the transcription of bacterial plasmids. I will identify the most significant gaps in our understanding, and suggest that the pursuit of a holistic model using integrative biology should improve our approaches to quantitative biology and engineering.

Andrea Scarpino, ICR

Computational Approaches for Targeted Protein Degradation in Cancer Drug Discovery

Targeted protein degradation has emerged as a transformative therapeutic strategy, offering the potential to eliminate traditionally “undruggable” proteins through proximity-induced ubiquitination. This presentation will explore the computational strategies used at the ICR Centre for Protein Degradation to support degrader design through cheminformatics, ligand-based and structure-based approaches. The design of a next-generation molecular glue degrader library and computational frameworks for ternary complex modelling will be discussed, highlighting the main challenges in the process. The integration of these computational methods will be shown to provide a rational foundation for designing targeted protein degraders with improved selectivity and therapeutic potential to help advance cancer drug discovery.

Benedict Irwin, Apheris

Federated Learning for Protein Co-Folding Using OpenFold3

We present an introduction to protein co-folding methods and the OpenFold3 model – an Open-Source AlphaFold3 Derivative. We then introduce federated learning and present early results on the federated training of the OpenFold3 model on both public and private data in the context of the AI Structural Biology Consortium (AISB) https://www.apheris.com/join-a-network/aisb.

Daniel Cole, Newcastle University

Open Science for Molecular Modelling with the Open Force Field Initiative

Drawing on computational methods that are based around training to extensive condensed phase physical property and quantum mechanical datasets, I will describe some of our efforts to design accurate and transferable inter- and intra-molecular potentials, with a view to applications in condensed phase atomistic modelling and computer-aided drug design.

I will explain how recent collaborations with the Open Force Field Initiative (https://openforcefield.org) enable the development of fast, accurate alternatives to traditional non-bonded functional forms. I will describe the development of a graph neural network based charge model targeting accurate electrostatic properties of organic molecules, and the use of Open Force Field infrastructure to train alternatives to the Lennard-Jones functional form. Finally, I will describe progress towards fast GPU-based optimisation of valence parameters for accurate simulation of conformational dynamics.

Daniela Dolciami and Rob Ziolek, Kvantify

Residence-time Scoring for High-Throughput Computer-Aided Drug Design

Accurate predictions of ligand residence times from molecular simulations typically come at a significant computational cost. This limits their deployment to low-throughput settings, which doesn’t suit typical drug discovery project requirements. We have developed a new simulation algorithm, Kvantify Koffee, that significantly accelerates the calculation of ligand residence times compared to existing approaches based on molecular dynamics simulations. Calculations are completed within minutes and are straightforward to set up, making our approach accessible even to non-expert users. Kvantify Koffee provides fast, reliable results that are generalizable across a diverse range of different systems without the training data requirements of machine learning models. Validation against a diverse set of experimentally characterized protein-ligand systems shows that our method predicts residence times with high accuracy, even for challenging non-congeneric ligand sets. Notably, our approach achieves these results with dramatically reduced computational costs, enabling the large-scale prioritization of compounds that would not be feasible using more computationally heavy methods. In this presentation, we will outline the theoretical foundation of our methodology, and highlight case studies showcasing practical applications of Kvantify Koffee to drug discovery tasks.

 

Jack Glancy, GSK

Active Learning FEP for Prospective Use in Drug Design

Free energy perturbation (FEP) methods are a powerful tool for protein-ligand affinity predictions but are limited to the order of 10s to 100s of compounds by GPU resource requirements. To cover a broader pool of enumerated ideas, machine learning/QSAR models built on FEP data can extend biochemical potency predictions to tens of thousands of compounds.

The process of iterative active learning to sample new FEP data points (AL-FEP) is an attractive proposition for improving the accuracy and domain of applicability of the QSAR models. To apply this method routinely to drug discovery projects, a retrospective evaluation of the AL-FEP workflow has been conducted covering parameters including the compound selection strategy, explore–exploit ratios, and number of compounds selected per cycle.

Significant differences in performance in terms of model enrichment and R2 are observed and rationalized. Recommendations are made as to when specific parameters should be employed for AL-FEP depending on the context (maximizing potency or broad-range prediction accuracy) in which the final model is to be deployed.

Pierre Matricon, Nxera Pharma

Modern Hit-Finding with Structure-Aware De Novo Design: Identification of Novel A2A Receptor Ligands Using Reinforcement Learning

G protein-coupled receptors (GPCRs) form a large group of flexible membrane proteins that are essential to numerous physiological processes and health conditions. Access to high-quality structural data is a cornerstone of our drug discovery efforts1,2. Our NxStaR™ technology has enabled the determination of over 400 unique GPCR structures, forming a robust foundation for our computational drug discovery projects.

This presentation will showcase our CADD strategies, from established physics-based methods3,4 to novel generative AI approaches5, used in the discovery and optimisation of A2A adenosine receptor (A2AR) antagonists. We will critically evaluate the performance of these methods and detail our innovative, multi-objective workflow that integrates de novo molecule generation with reinforcement learning. This approach allows for the efficient and targeted exploration of vast chemical space, which is critical for identifying novel hits for well-characterized targets such as A2A. This multi-objective, structure-guided scoring strategy successfully guided the design and the synthesis of nine proposed molecules. Five of them exhibited varied modalities against the A2AR, with two potent inverse agonists which yielded high resolution crystal structures5. This outcome validates our computational methodology and highlights the power of combining experimental structural biology with advanced computational design to overcome SBDD challenges.

References:

[1] M. Congreve, C. de Graaf, N.A. Swain, C.G. Tate, Cell, (2020) 181. DOI: 10.1016/j.cell.2020.03.003.

[2] F. Ballante, A.J. Kooistra, S. Kampen, C. de Graaf, J. Carlsson, Pharm Rev, (2021) 73, 4. DOI: 10.1124/pharmrev.120.000246.

[3] M. Congreve, S.P. Andrews, A.S. Doré , K. Hollenstein, E. Hurrell, C.J. Langmead, J.S. Mason, I.W. Ng, B. Tehan, A. Zhukov, M. Weir, F.H. Marshall, J Med Chem, (2012) 55. DOI: 10.1021/jm201376w.

[4] F. Deflorian, L. Perez-Benito, E.B. Lenselink, M. Congreve, H.W.T. van Vlijmen, J.S. Mason, C. de Graaf, G. Tresadern, J Chem Inf Model, (2020), 60, 11. DOI: 10.1021/acs.jcim.0c00449.

[5] M. Thomas, P.G. Matricon, R.J. Gillespie, M. Napiórkowska, H. Neale, J.S. Mason, J. Brown, K. Harwood, C. Fieldhouse, N.A. Swain, T. Geng, N.M. O’Boyle, F. Deflorian, A. Bender, C. de Graaf, Nat Commun, (2025) 16. DOI: 10.1038/s41467-025-60629-0.


Jobs


Upcoming Meetings

Nessa Carson of AZ has been compiling a comprehensive list of scientific conferences, which can be found here.

Notable conferences of particular interest to our readers include:

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.