Biology subjects

Deane, C.

Publications and source records attributed to Deane, C..

At least 19 recordsLinked to original sources

bioRxiv2026

Generative Language Modeling for Antibody CDR Grafting and Alignment-driven De Novo Design

Antibodies recognise their targets through hypervariable complementarity-determining regions (CDRs), which are interleaved with conserved frameworks in sequence space, making de novo CDR design an infilling problem. Autoregressive models generate residues left-to-right, which precludes full framework context during CDR generation and conflates framework and CDR likelihoods, leaving no natural prompt-response interface for feedback to steer generation. We present GenCDR, a family of LLaMa-based autoregressive language models that read all frameworks as a conditioning prompt and generate all CDRs jointly as a variable-length response, making CDR likelihoods a clean, separable target for reward attribution. The family comprises IgGenCDR, p-IgGenCDR, and NanoGenCDR, trained on unpaired, paired, and nanobody chains, respectively. GenCDR achieves the highest CDR recovery among autoregressive models and produces natural, diverse, human-like CDRs whose likelihoods correlate with fitness and developability assays. The prompt-response boundary also enables principled alignment: reward signals for binding affinity, expression, or developability can be composed to steer CDR generation. Over four rounds of alignment against antibody-antigen co-folding and developability objectives, we find that NanoGenCDR, which uses no explicit antigen encoding, can reach in silico structural interface metrics competitive with those of a structure-conditioned diffusion pipeline at roughly half the sampling budget, with more natural, developable designs. The same interface can be extended to integrate experimental feedback, opening a path to closed-loop antibody de novo design.

synthetic biology↗

bioRxiv2026

Characterising AlphaFold 3s ability to predict T cellantigen specificity

T cells are a key part of the adaptive immune system. Using their surface-bound T cell antigen receptors (TCRs), these cells scan peptides and other antigens presented to them by major histocompatibility complex molecules (MHCs) on the surface of cells, searching for abnormalities. Although determining the map between TCRs and their target antigens is of vital importance for the design of safe and effective T cell-based vaccines and therapeutics, decoding these interactions is challenging. Experimental methods are not scalable, and sequence-based computational methods have issues generalising to new antigens. The IMMREP25 benchmark of methods for predicting T cell antigen specificity showed that AlphaFold-based methods promise improved generalisation to novel antigens. However, the ability of structure prediction models to predict T cell antigen specificity has not been robustly evaluated previously. In this work, we characterise AlphaFolds ability to predict T cell antigen specificity. We created a pipeline for high-throughput prediction of TCR:peptide-MHC (pMHC) structures using AlphaFold that is > 100 fold faster than the default implementation and used it to benchmark AlphaFold 3 (AF3) and similar models at predicting T cell antigen specificity. We investigated the underlying correlates of AlphaFold-derived binding scores and found that the models predictive power is related to the positioning of TCRs over the pMHC and not chemical interactions. Furthermore, we refine the AlphaFold-derived binding scores by training a machine learning model we call the PAE Aggregator. We then investigate AF3s ability to uncover the clustering rules of TCR repertoires and recapitulate mutational scanning experiments. These analyses show that AlphaFold3 clusters sequence-similar TCRs according to their binding mode and detects disrupting point mutations accurately. Our results highlight both the promise and the current limitations of structure-based approaches for predicting TCR specificity, guiding the development of more reliable immunological prediction methods.

systems biology↗

bioRxiv2026

Just Add Structure: Protein Language Models Combined with Structural Equivariance Excel at Protein Tasks

Accurate in silico prediction of protein properties, functional fitness, and mutational effects remains a central challenge in protein engineering and therapeutic design. While Protein Language Models (PLMs) successfully capture rich evolutionary and functional constraints from sequence data, they only indirectly encode the spatial and geometric information that fundamentally governs protein function. Consequently, state-of-the-art approaches typically rely on extensive fine-tuning, ensembling, or the incorporation of handcrafted structural features to achieve competitive accuracy, making them computationally expensive and difficult to scale. In this work, we demonstrate that explicit geometric modeling can substitute for, and in most cases outperform, large-scale PLM fine-tuning, with much higher parameter efficiency. Our approach, ProtEGNN, pairs PLM residue representations with a lightweight E(3)-Equivariant Graph Neural Network, competing with or achieving state-of-the-art performance across eight different benchmarks in protein property, mutational effect and function prediction, while needing 100-1000x fewer parameters than competing methods. Even when protein structure is combined with representations from ESM2-T6, a small 8M-parameter PLM, ProtEGNN matches fine-tuned sequence-only approaches based on substantially larger PLM backbones, while training orders of magnitude fewer parameters. Together, these results highlight geometric inductive bias as a powerful and scalable alternative to task-specific fine-tuning of large PLMs for protein modeling.

bioinformatics↗

bioRxiv2026

A framework for testing structural hypotheses of protein dynamics against experimental HDX-MS data

Protein dynamics determine biological function, yet extracting structural ensembles from Hydrogen-Deuterium Exchange Mass Spectrometry (HDX-MS) remains a challenging inverse problem. Current ensemble-fitting approaches often achieve good agreement with uptake curves but lack rigorous validation and uncertainty quantification, limiting structural confidence. We propose ValDX, a validation framework for quantitative integration of HDX-MS data with structural ensembles. ValDX combines overlap-aware data splitting, replicate-based uncertainty estimation, and uptake-independent "Work Done" metrics that quantify how much an ensemble must be modified to match experiment. Across 22 ensembles spanning six proteins (58-474 residues), we show that conventional error metrics fail to distinguish structurally representative ensembles from incorrect ones, whereas Work Done metrics robustly discriminate global and local conformational quality. We further demonstrate that clustering yields compact, interpretable ensembles with minimal loss of accuracy, and that staged optimisation enables reliable fitting of both ensemble weights and forward-model parameters without requiring a reference structure. Together, this framework establishes HDX-MS ensemble integration as a quantitative structural hypothesis-testing problem, enabling inference of protein dynamics from HDX-MS data.

biophysics↗

bioRxiv2025

INTERPOLATION-BASED CONDITIONING OF FLOW MATCHING MODELS FOR BIOISOSTERIC LIGAND DESIGN

AO_SCPLOWBSTRACTC_SCPLOWFast, unconditional 3D generative models can now produce high-quality molecules, but adapting them for specific design tasks often requires costly retraining. To address this, we introduce two training-free, inference-time conditioning strategies, Interpolate-Integrate and Replacement Guidance, that provide control over E(3)-equivariant flow-matching models. Our methods generate bioisosteric 3D molecules by conditioning on seed ligands or fragment sets to preserve key determinants like shape and pharmacophore patterns, without requiring the original fragment atoms to be present. We demonstrate their effectiveness on three drug-relevant tasks: natural product ligand hopping, bioisosteric fragment merging, and pharmacophore merging.

molecular biology↗

bioRxiv2025

AntiDIF: Accurate and Diverse Antibody Specific Inverse Folding with Discrete Diffusion

Inverse folding is an important step in current computational antibody design. Recently deep learning methods have made impressive progress in improving the sequence recovery of antibodies given their 3D backbone structure. However, inverse folding is often a one-to-many problem, i.e. there are multiple sequences that fold into the same structure. Previous methods have not taken into account the diversity between the predicted sequences for a given structure. Here we create AntiDIF an Antibody-specific discrete Diffusion model for Inverse Folding. Compared with stateof-the-art methods we show that AntiDIF improves diversity between predictions while keeping high sequence recovery rates. Furthermore, forward folding of the generated sequences shows good agreement with the target 3D structure.

immunology↗

bioRxiv2025

AF2χ: Predicting protein side-chain rotamer distributions with AlphaFold2

The flexibility of protein side chains is an essential contributor of conformational entropy and affects processes such as folding, stability and molecular interactions. Structure determination experiments and prediction tools such as AlphaFold generally fail to capture or represent the conformational heterogeneity of proteins in solution. Experiments can be used to study side-chain flexibility, but cannot be applied at scale, and most prediction methods focus on reconstructing the minimum free energy state rather than an ensemble representing side-chain configurations. Here, we use AlphaFold2 and its internal side-chain representations to develop AF2{chi} that predicts side-chain {chi}-angle distributions and generates structural ensembles. We extensively benchmark AF2{chi} predictions using experimental NMR 3J-couplings and s2 order parameters, as well as dihedral angle distributions derived from collections of experimental structures, demonstrating the accuracy of AF2{chi} in generating accurate side-chain ensembles. We also compare the accuracy of AF2{chi} with molecular dynamics simulations and recent machine learning models aimed to generate conformational ensembles and show that AF2{chi} provides state-of-the-art accuracy orders of magnitude faster than molecular simulations. With its speed and accuracy, AF2{chi} offers a strong complementary option to simulations and rotamer library approaches, making it particularly valuable for applications such as protein design, ligand docking and interpretation of biophysical experiments.

biophysics↗

bioRxiv2025

Supervised Deep Learning for Efficient Cryo-EM Image Alignment in Drug Discovery with cryoPARES

Cryo-Electron Microscopy (cryo-EM) is a pivotal tool for determining 3D structures of biological macromolecules. Current workflows are computationally demanding and require manual intervention, creating bottlenecks for high-throughput applications like structure-based drug discovery. In such contexts, where all protein samples can be assumed to be equivalent at resolutions relevant for image alignment, information about particle poses from previous refinements could be reused. Existing methods, however, ignore this prior knowledge, aligning each dataset from scratch. We present cryoPARES, a deep learning pose estimation method trained on pre-aligned datasets. Our method not only provides accurate angular predictions significantly faster than traditional approaches but also introduces automated particle pruning capabilities that eliminate manual intervention. Together with its single-pass operation, these features enable near real-time reconstructions that provide feedback during data acquisition. We demonstrate cryoPARESs effectiveness through rapid structural determination of seven ligand-bound complexes across four distinct protein targets. We also release three fragment-bound cryo-EM datasets.

bioinformatics↗

bioRxiv2024

Quantifying conformational changes in the TCR:pMHC-I binding interface

T cells form one of the key pillars of adaptive immunity. Using their surface bound T cell antigen receptors (TCRs), these cells screen millions of antigens presented by major histocompatability complex (MHC) or MHC-like molecules. In other protein families, the dynamics of protein-protein interactions have important implications for protein function. Case studies of TCR:class I peptide-MHCs (pMHC-Is) structures have reported mixed results on whether the binding interfaces undergo conformational change during engagement and no robust statistical quantification has been done to generalise these results. It thus remains an open question if movement occurs in the binding interface that enables recognition and activation of T cells. In this work, we quantify the conformational changes in the TCR:pMHC-I binding interface by creating a dataset of 358 structures, comprising 25 TCRs, 20 MHC alleles, and 58 peptide structures in both unbound (apo) and bound (holo) conformations. In support of some case studies, we demonstrate that all complimentary determining region (CDR) loops move to a certain extent but only CDR3 and CDR3{beta} loops modify their shape when binding pMHC-Is. We also map out the contacts between TCRs and pMHC-Is, generating a novel fingerprint of TCRs on MHC molecules and show that the CDR3 tends to bind the N-terminus of the peptide and the CDR3{beta} tends to bind the C-terminus of the peptide. Finally, we show that the presented peptides can undergo conformational changes when engaged by TCRs, as has been reported in past literature, but novelly show these changes depend on how the peptides are anchored in the MHC binding groove. Our work has implications in understanding the behaviour of TCR:pMHC-I interactions and providing insights that can be used for modelling T cell antigen specificity, an ongoing grand challenge in immunology.

immunology↗

bioRxiv2024

Assessing AF2's ability to predict structural ensembles of proteins

Recent breakthroughs in protein structure prediction have enhanced the precision and speed at which protein configurations can be determined, setting new benchmarks for accuracy and efficiency in the field. However, the fundamental mechanisms of biological processes at a molecular level are often connected to conformational changes of proteins. Molecular dynamics (MD) simulations serve as a crucial tool for capturing the conformational space of proteins, providing valuable insights into their structural fluctuations. However, the scope of MD simulations is often limited by the accessible timescales and the computational resources available, posing challenges to comprehensively exploring protein behaviors. Recently emerging approaches have focused on expanding the capability of AlphaFold2 (AF2) to predict conformational substates of protein structures by manipulating the input multiple sequence alignment (MSA). These approaches operate under the assumption that the MSA also contains information about the heterogeneity of protein structures. Here, we benchmark the performance of various workflows that have adapted AF2 for ensemble prediction focusing on the subsampling of the MSA as implemented in ColabFold and compare the obtained structures with ensembles obtained from MD simulations and NMR. As test cases, we chose four proteins namely the bovine pancreatic inhibitor protein (BPTI), thrombin and two antigen binding fragments (antibody Fv and nanobody), for which reliable experimentally validated structural information (X-ray and/or NMR) was available. Thus, we provide an overview of the levels of performance and accessible timescales that can currently be achieved with machine learning (ML) based ensemble generation. In three out of the four test cases, we find structural variations fall within the predicted ensembles. Nevertheless, significant minima of the free energy surfaces remain undetected. This study highlights the possibilities and pitfalls when generating ensembles with AF2 and thus may guide the development of future tools while informing upon the results of currently available applications.

biophysics↗

bioRxiv2024

Inferring residue level hydrogen deuterium exchange with ReX

Hydrogen-Deuterium Exchange Mass-Spectrometry (HDX-MS) has emerged as a powerful technique to explore the conformational dynamics of proteins and protein complexes in solution. The bottom-up approach to MS uses peptides to represent an average of residues, leading to reduced resolution of deuterium exchange and complicates the interpretation of the data. Here, we introduce ReX, a method to infer residue-level uptake patterns leveraging the overlap in peptides, the temporal component of the data and the correlation along the sequence dimension. This approach infers statistical significance for individual residues by treating HDX-MS as a multiple change-point problem. By fitting our model in a Bayesian non-parametric framework, we perform parameter number inference, differential HDX confidence assessments, and uncertainty estimation for temporal kinetics. Benchmarking against existing methods using a three-way proteolytic digestion experiment shows our methods superior performance at predicting unseen HDX data. Moreover, it aligns HDX-MS with the reporting standards of other structural methods by providing global and local resolution metrics. Using ReX, we analyze the differential flexibility of BRD4s two Bromodomains in the presence of I-BET151 and quantify the conformational variations induced by a panel of seventeen small molecules on LXR. Our analysis reveals distinct residue-level HDX signatures for ligands with varied functional outcomes, highlighting the potential of this characterisation to inform mode of action analysis.

biochemistry↗

bioRxiv2024

MolSnapper: Conditioning Diffusion for Structure Based Drug Design

Generative models have emerged as potentially powerful methods for molecular design, yet challenges persist in generating molecules that effectively bind to the intended target. The ability to control the design process and incorporate prior knowledge would be highly beneficial for better tailoring molecules to fit specific binding sites. In this paper, we introduce MolSnapper, a novel tool that is able to condition diffusion models for structure-based drug design by seamlessly integrating expert knowledge in the form of 3D pharmacophores. We demonstrate through comprehensive testing on both the CrossDocked and Binding MOAD datasets, that our method generates molecules better tailored to fit a given binding site, achieving high structural and chemical similarity to the original molecules. It also, when compared to alternative methods, yields approximately twice as many valid molecules.

bioinformatics↗

bioRxiv2024

Addressing the antibody germline bias and its effect on language models for improved antibody design

AO_SCPLOWBSTRACTC_SCPLOWThe versatile binding properties of antibodies have made them an extremely important class of biotherapeutics. However, therapeutic antibody development is a complex, expensive and time-consuming task, with the final antibody needing to not only have strong and specific binding, but also be minimally impacted by any developability issues. The success of transformer-based language models in protein sequence space and the availability of vast amounts of antibody sequences, has led to the development of many antibody-specific language models to help guide antibody discovery and design. Antibody diversity primarily arises from V(D)J recombination, mutations within the CDRs, and/or from a small number of mutations away from the germline outside the CDRs. Consequently, a significant portion of the variable domain of all natural antibody sequences remains germline. This affects the pre-training of antibody-specific language models, where this facet of the sequence data introduces a prevailing bias towards germline residues. This poses a challenge, as mutations away from the germline are often vital for generating specific and potent binding to a target, meaning that language models need be able to suggest key mutations away from germline. In this study, we explore the implications of the germline bias, examining its impact on both general-protein and antibody-specific language models. We develop and train a series of new antibody-specific language models optimised for predicting non-germline residues. We then compare our final model, AbLang-2, with current models and show how it suggests a diverse set of valid mutations with high cumulative probability. AbLang-2 is trained on both unpaired and paired data, and is freely available (https://github.com/oxpig/AbLang2.git).

bioinformatics↗

bioRxiv2023

Towards the accurate modelling of antibody-antigen complexes from sequence using machine learning and information-driven docking

Antibody-antigen complex modelling is an important step in computational workflows for therapeutic antibody design. While experimentally determined structures of both antibody and the cognate antigen are often not available, recent advances in machine learning-driven protein modelling have enabled accurate prediction of both antibody and antigen structures. Here, we analyse the ability of protein-protein docking tools to use machine learning generated input structures for information-driven docking. We find that HADDOCK can generate accurate models of antibodyantigen complexes using an ensemble of antibody structures generated by machine learning tools and AlphaFold2 predicted antigen structures. Targeted docking using knowledge of the complementary determining regions on the antibody and some information about the targeted epitope allows the generation of high quality models of the complex with reduced sampling, resulting in a computationally cheap protocol that outperforms the ZDOCK baseline. The data set used to benchmark the docking protocols in this study is available at github.com/haddocking/ai-antibodies. The docking models will be deposited at data.sbgrid.org/labs/32/ upon acceptance.

bioinformatics↗

bioRxiv2023

Robustly interrogating machine learning based scoring functions: what are they learning?

MotivationMachine learning-based scoring functions (MLBSFs) have been found to exhibit inconsistent performance on different benchmarks and be prone to learning dataset bias. For the field to develop MLBSFs that learn a generalisable understanding of physics, a more rigorous understanding of how they perform is required. ResultsIn this work, we compared the performance of a diverse set of popular MLBSFs (RFScore, SIGN, OnionNet-2, Pafnucy, and PointVS) to our proposed baseline models that can only learn dataset biases on a range of benchmarks. We found that these baseline models were competitive in accuracy to these MLBSFs in almost all proposed benchmarks, indicating these models only learn dataset biases. Our tests and provided platform, ToolBoxSF, will enable researchers to robustly interrogate MLBSF performance and determine the effect of dataset biases on their predictions. Availability and Implementationhttps://github.com/guydurant/toolboxsf Contactdeane@stats.ox.ac.uk Supplementary informationSupplementary data are available at Bioinformatics online.

bioinformatics↗

bioRxiv2023

The molecular reach of antibodies determines their SARS-CoV-2 neutralisation potency

Key functions of antibodies, such as viral neutralisation, depend on bivalent binding but the factors that influence it remain poorly characterised. Here, we develop and employ a new bivalent model to mechanistically analyse binding between >45 patient-isolated IgG1 antibodies interacting with SARS-CoV-2 RBD surfaces. Our method reproduces the monovalent on/off-rates and enables measurements of the bivalent on-rate and the molecular reach: the maximum antigen separation that supports bivalent binding. We find large variations in these parameters across antibodies, including variations in reach (22-46 nm) that exceed the physical antibody size ([~]15 nm) due to the antigen size. The bivalent model integrates all parameters, including reach and antigen density, to predict an emergent binding potency for each antibody that matches their neutralisation potency. Indeed, antibodies with similar monovalent affinities to the same RBD-epitope but with different reaches display differences in emergent bivalent binding that match differences in their neutralisation potency. Together, our work highlights that antibodies within an isotype class binding the same antigen can display differences in molecular reach that can substantially modulate their emergent binding and functional properties. Lay SummaryAntibodies are soluble proteins that can neutralise pathogens by sticking to them. They contain two identical arms that allow them to simultaneously bind two identical antigen molecules on pathogen surfaces. Although we know that bivalent binding is important for neutralisation, we dont know how different antibodies achieve it. We developed a new model to analyse the mechanism of bivalent binding and used it to study over 45 antibodies from COVID-19 patients that bind the RBD antigen of SARS-CoV-2. Unexpectedly, we found that the molecular reach of an antibody, which is the maximum antigen separation that supports bivalent binding, varied widely between antibodies and exceeded their physical size. We show how antibody binding emerges from the interplay of multiple factors, including reach, and that this emergent binding predicts their neutralisation function. The ability to analyse and predict bivalent binding should improve our understanding and exploitation of antibodies. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=155 SRC="FIGDIR/small/556503v2_ufig1.gif" ALT="Figure 1"> View larger version (34K): org.highwire.dtl.DTLVardef@1674563org.highwire.dtl.DTLVardef@fc7feborg.highwire.dtl.DTLVardef@1325847org.highwire.dtl.DTLVardef@41ad51_HPS_FORMAT_FIGEXP M_FIG C_FIG

immunology↗

bioRxiv2023

PEP-Patch: Electrostatics in Protein-Protein Recognition, Specificity and Antibody Developability

The electrostatic properties of proteins arise from the number and distribution of polar and charged residues. Due to their long-ranged nature, electrostatic interactions in proteins play a critical role in numerous processes, such as molecular recognition, protein solubility, viscosity, and antibody developability. Thus, characterizing and quantifying electrostatic properties of a protein is a pre-requisite for understanding these processes. Here, we present PEP-Patch, a tool to visualize and quantify the electrostatic potential on the protein surface and showcase its applicability to elucidate protease substrate specificity, antibody-antigen recognition and predict heparin column retention times of antibodies as an indicator of pharmacokinetics.

biophysics↗

bioRxiv2023

Microfluidics-enabled fluorescence-activated cell sorting of single pathogen-specific antibody secreting cells for the rapid discovery of monoclonal antibodies

Monoclonal antibodies are increasingly used to prevent and treat viral infections, playing a pivotal role in pandemic response efforts. Antibody secreting cells (ASCs, plasma cells and plasmablasts) are an excellent source of high-affinity antibodies with therapeutic potential. Current methodologies to study antigen-specific ASCs either have low throughput, require expensive and labour-intensive screening or are technically demanding and therefore not accessible to the wider research community. Here, we present a straightforward technology for the rapid discovery of monoclonal antibodies from ASCs: we combine microfluidic encapsulation of single cells into an antibody capture hydrogel with antigen bait sorting by conventional flow cytometry. With our technology, we screened millions of mouse and human ASCs and obtained anti-SARS-CoV-2 monoclonal antibodies with high affinity (pM) and neutralising capacity (<100 ng/mL) in two weeks with a high hit rate (>85%). By facilitating access into the underexplored ASC compartment, we enable fast and efficient antibody discovery as well as immunological studies into the generation of protective antibodies.

immunology↗