Biology subjects

Bryant, P.

Publications and source records attributed to Bryant, P..

18 recordsLinked to original sources

bioRxiv2026

De novo Design of Macrocyclic Molecular Glues

The engineering of induced proximity has transformed drug discovery, yet the development of molecular glues remains largely serendipitous and restricted to the retrospective optimisation of accidental discoveries. Here, we present EvoBind-multimer, a deep learning framework for the de novo design of molecular glues directly from protein sequences. Unlike structure-based docking, our method generates small macrocyclic peptides that bridge user-defined protein pairs without requiring prior interface knowledge or existing ligands. We applied this framework to recruit the E3 ligase VHL to two challenging oncoproteins: KRAS and BRD4. Live-cell NanoBRET demonstrated robust design-induced proximity for both pairs. Mechanistic validation demonstrated that the generated macrocycles form functional VHL-target ternary complexes capable of driving Cullin-RING ligase-dependent proteasomal degradation and downstream signalling shutdown. Finally, evaluation in patient-derived xenograft neuroblastoma tumoroids revealed that ternary complex processing is deeply context-dependent: identical macrocycles acted as potent degraders in one patient model, yet functioned as stabilising "LOCKTACs" in another, driving VHL-dependent target sequestration without turnover. By enabling the de novo design of induced proximity from sequence alone, EvoBind-multimer provides a route towards designing new protein functions.

bioinformatics↗

bioRxiv2026

The Structural History of Eukarya

Comparative genomics has traditionally relied on sequence-based markers, yet the global evolutionary landscape of protein architecture remains largely unexplored. We present the Structural History of Eukarya (SHE), a phylogeny derived from all-vs-all structural comparisons of 1,542 eukaryotic proteomes, encompassing nearly 300 trillion protein-protein alignments. This proteome-scale analysis reveals a bipartite model of eukaryotic evolution: a rigid Strict Core dominated by cytoskeletal architecture, supporting a more plastic Operational Engine centred on translational machinery. We identify lineage-specific accelerations of structural evolution in species such as birds and ants that are decoupled from proteome expansion, and show that aggregate structural topology provides a quantitative diagnostic of reference proteome quality. Finally, we demonstrate that proteome-wide structural fidelity enables a data-driven framework for model organism selection, replacing heuristic choices with quantitative matching to human biological processes. SHE is freely accessible as an interactive portal: https://she-app.serve.scilifelab.se/

evolutionary biology↗

bioRxiv2025

RareFoldGPCR: Agonist Design Beyond Natural Amino Acids

Noncanonical amino acids (NCAAs) expand the chemical diversity of peptides beyond the twenty standard residues, offering new opportunities for designing binders with novel interaction modes and functional activity. G protein-coupled receptors (GPCRs) are central to cellular signalling and represent one of the largest classes of therapeutic targets, yet their functional modulation remains challenging. Here, we present RareFoldGPCR (RFG), a GPCR-specialised AI model for structure prediction and design that supports NCAAs. By applying transfer learning on high-resolution GPCR structures, RFG accurately models and rationally designs both linear and cyclic peptides that incorporate NCAAs and modulate GPCR activity. This is achieved without the model ever being trained on NCAA-based GPCR modulators. We showcase the capability of RFG by designing peptide agonists for the glucagon-like peptide-1 receptor (GLP1R) and validating their functional activity experimentally in cell-based assays. We investigate the precise capabilities of generating active agonists by expanding different regions of the native GLP-1 hormone, and further demonstrate the design of cyclic peptide agonists with entirely novel sequences and topologies, creating new agonist modes. We analyse how design metrics relate to pathway specificity, enabling precise modulation of pathway activity, such as activating the cAMP response without recruiting {beta}-arrestin to reduce receptor desensitisation. RFG shows how transfer learning on specific target classes enables generalisation to new chemistry and molecular topology, providing a broadly applicable strategy for designing functional ligands beyond the constraints of natural amino acid chemistry. RFG is freely available: https://github.com/patrickbryant1/RareFoldGPCR

bioinformatics↗

bioRxiv2025

Blind De Novo Design of Dual Cyclic Peptide Agonists Targeting GCGR and GLP1R

Agonists of the glucagon receptor (GCGR) and glucagon-like peptide-1 receptor (GLP1R) are key to treating metabolic diseases like type 2 diabetes and obesity, but often limited by being linear. Cyclic peptides offer greater stability and potential oral delivery, but their de novo design with computational methods as agonists is unproven. Here, we used the EvoBind AI platform to design cyclic peptide agonists for GCGR from sequence alone, without any structural templates or prior binding site information. EvoBind consistently identified the receptors activation surface, producing peptides that also activated GLP1R. Of three synthesised peptides, two showed potent dual activation of GCGR and GLP1R in cAMP assays, while all three activate GLP1R, with the 18-residue design achieving EC50 values of 32 nM (GLP1R) and 542 nM (GCGR), rivalling natural hormones. The designed sequences show no similarity to known agonists, and structural modelling indicates that they adopt novel binding modes compatible with active-state stabilisation. This is made possible by EvoBinds purely sequence-based, template-free approach, which enables the discovery of alternative solutions that are inaccessible to structure-guided methods. This study presents the first successful blind de novo design of a cyclic peptide agonist for a G protein-coupled receptor (GPCR), using only the receptor sequence as input. The findings highlight EvoBinds capacity to go beyond static binding, offering a generalisable strategy for therapeutic discovery through target-agnostic, sequence-driven peptide design.

bioinformatics↗

bioRxiv2025

RareFold: Structure prediction and design of proteins with noncanonical amino acids

Protein structure prediction and design have traditionally been confined to the 20 canonical amino acids. Expanding this chemical space to include non-canonical amino acids (ncAAs) is essential for engineering proteins with novel chemical and functional properties. However, existing methods are not designed to generalise across chemically diverse residue types. Here, we present RareFold, a deep learning architecture for structure prediction and design of proteins containing the 20 canonical amino acids and 29 ncAAs. By representing each residue as an independent token, RareFold learns context-dependent atomic interaction patterns across chemically diverse sequence spaces, enabling modelling of non-standard chemistries within a unified framework. We apply this capability in EvoBindRare, a generative framework for de novo design of linear and cyclic peptide binders with an efficient implementation that substantially reduces computational requirements compared to existing architectures. We demonstrate its performance by designing binders against Ribonuclease A, yielding novel linear and cyclic peptides incorporating ncAAs within predicted interfaces with low-micromolar affinities (KD [~]2-9 M), comparable to the native ligand (KD [~]2 M). Hydrogen-deuterium exchange mass spectrometry confirms that the designed peptides engage the target at regions consistent with predicted binding interfaces. In addition, immunogenicity profiling in human-derived organoid models shows no detectable immune activation. By extending deep learning-based protein design to non-canonical chemical spaces, RareFold enables programmable access to expanded amino acid alphabets and broadens the scope of de novo protein engineering.

bioinformatics↗

bioRxiv2025

Single-Shot Design of a Cyclic Peptide Inhibitor of HIV-1 Membrane Fusion with EvoBind

HIV evades the immune system through rapid mutation of its surface proteins, particularly the envelope glycoprotein. However, the core mechanism of viral entry, CD4 binding and co-receptor engagement remains conserved. While therapies such as Lenacapavir represent important advances, the continued emergence of resistant strains will demand new and more adaptable treatment strategies. This challenge is not unique to HIV; future pandemics will likely present similar pressures, highlighting the need for drug design methods that are not only effective but also fast and scalable. Recent advances in protein structure prediction have transformed the landscape of therapeutic design, enabling the accurate modelling of target structures from sequence alone and now facilitating the development of novel therapeutics without prior structural data. EvoBind leverages these advances to rapidly generate cyclic peptide binders in a single design round, using only the amino acid sequence of a target protein. Cyclic peptides offer several advantages over traditional linear protein molecules, including increased stability, while their small size improves oral bioavailability and enables access to challenging binding sites. Here, we demonstrate the use of EvoBind to generate cyclic peptide binders against the HIV envelope protein gp41, which is essential for viral-host membrane fusion. Cell-based assays confirm potent inhibition of two different HIV-1 strains with no detectable toxicity. The combination of artificial intelligence-guided design and streamlined experimental validation can significantly accelerate therapeutic development, reduce costs, and provide timely solutions to the challenges posed by viral evolution and emerging global health threats.

pharmacology and toxicology↗

bioRxiv2024

AI-first structural identification of pathogenic protein targets

The likelihood for pandemics is increasing as the world population grows and becomes more interconnected. Obtaining structural knowledge of protein-protein interactions between a pathogen and its host can inform pathogenic mechanisms and treatment or vaccine design. Currently, there are 52 nonredundant human-pathogen interactions with known structure in the PDB, although there are 21064 with experimental support in the HPIDB, meaning that only 0.2% of known interactions have known structure. Recent improvements in structure prediction of protein complexes based on AlphaFold have made it possible to model heterodimeric complexes with very high accuracy. However, it is not known how this translates to host-pathogen interactions which share a different evolutionary relationship. Here, we analyse the structural protein-protein interaction network between ten different pathogens and their human host. We predict the structure of 9452 human-pathogen interactions of which only 10 have known structure. We find that we can model 30 interactions with an expected TM-score of [≥]0.9, expanding the structural knowledge in these networks three-fold. We select the highly-scoring Francisella tularensis dihydroprolyl dehydrogenase (IPD) complex with human immunoglobulin Kappa constant (IGKC) for detailed analysis with homology modeling and native mass spectrometry. Our results confirm the predicted 1:2:1 heterotetrameric complex with potential implications for bacterial immune response evasion. We are entering a new era where structure prediction can be used to guide vaccine and drug development towards new pathogenic targets in very short time frames.

bioinformatics↗

bioRxiv2024

Design of linear and cyclic peptide binders of different lengths only from a protein target sequence

Structure prediction technology has revolutionised the field of protein design, but key questions such as how to design new functions remain. Many proteins exert their functions through interactions with other proteins, and a significant challenge is designing these interactions effectively. While most efforts have focused on larger, more stable proteins, shorter peptides offer advantages such as lower manufacturing costs, reduced steric hindrance, and the ability to traverse cell membranes when cyclized. However, less structural data is available for peptides and their flexibility makes them harder to design. Here, we present a method to design both novel linear and cyclic peptide binders of varying lengths based solely on a protein target sequence. Our approach does not specify a binding site or the length of the binder, making the procedure completely blind. We demonstrate that linear and cyclic peptide binders of different lengths can be designed with nM affinity in a single shot, and adversarial designs can be avoided through orthogonal in silico evaluation, tripling the success rate. Our protocol, EvoBind2 is freely available https://github.com/patrickbryant1/EvoBind.

bioinformatics↗

bioRxiv2023

Structure prediction of protein-ligand complexes from sequence information with Umol

Protein-ligand docking is an established tool in drug discovery and development to narrow down potential therapeutics for experimental testing. However, a high-quality protein structure is required and often the protein is treated as fully or partially rigid. Here we develop an AI system that can predict the fully flexible all-atom structure of protein-ligand complexes directly, given a multiple sequence alignment representation of the protein and a SMILES string representing the ligand. At a high accuracy threshold, unseen protein-ligand complexes can be predicted more accurately than for RoseTTAFold-AA, and at medium accuracy even classical docking methods that use known protein structures as input are surpassed. The high accuracy presented here suggests that the goal of AI-based drug discovery is one step closer, but there is still a way to go to fully grasp the complexity of protein-ligand interactions. Umol is available at: https://github.com/patrickbryant1/Umol

bioinformatics↗

bioRxiv2023

Structure prediction of alternative protein conformations

Proteins are dynamic molecules whose movements result in different conformations with different functions. Neural networks such as AlphaFold2 can predict the structure of single-chain proteins in the conformations most likely to exist in the PDB. However, almost all conformations in the PDB are seen during training. Therefore, it is not possible to assess whether alternative protein conformations can be predicted or if these are reproduced from memory. Here, we train a new structure prediction network on a conformational split of the PDB to generate alternative conformations. Our network, Cfold, enables efficient exploration of the conformational landscape of monomeric protein structures. 52% (81) of the nonredundant alternative protein conformations evaluated here are predicted with high accuracy (TM-score>0.8). Cfold is freely available at: https://github.com/patrickbryant1/Cfold

bioinformatics↗

bioRxiv2023

Improved protein complex prediction with AlphaFold-multimer by denoising the MSA profile

Structure prediction of protein complexes has improved significantly with AlphaFold2 and AlphaFold-multimer (AFM), but only 60% of dimers are accurately predicted. A way to improve the predictions is to inject noise to generate more diverse predictions. However, thousands of predictions are needed to obtain a few that are accurate in difficult cases. Here, we learn a bias to the MSA representation that improves the predictions by performing gradient descent through the AFM network. We effectively denoise the MSA profile, similar to how a blurry image would be sharpened. We demonstrate the performance on seven difficult targets from CASP15 and increase the average MMscore to 0.76 compared to 0.63 with AFM. We evaluate the procedure on 334 protein complexes where AFM fails and demonstrate an increased success rate (MMscore>0.75) of 8% on these hard targets. Our protocol, AFProfile, provides a way to direct predictions towards a defined target function guided by the MSA. We expect gradient descent over the MSA to be useful for different tasks, such as generating alternative conformations. AFProfile is freely available at: https://github.com/patrickbryant1/AFProfile

bioinformatics↗

bioRxiv2023

Rapid protein-protein interaction network creation from multiple sequence alignments with Deep Learning

AlphaFold2 (AF) can evaluate protein-protein interactions (PPIs) with high accuracy by finding evolutionary signals between proteins but comes with a high computational cost. Here, we speed up the prediction with AF for PPI network prediction 40x and reduce the disk space requirements 4000x for a set of 1000 proteins. Our protocol is easy to install and freely available from: https://github.com/patrickbryant1/SpeedPPI.

bioinformatics↗

bioRxiv2022

EvoBind: in silico directed evolution of peptide binders with AlphaFold

Currently, there is no accurate method to computationally design peptide binders towards a specific protein interface using only a target structure. Experimental methods such as phage display can produce strong binders, but it is impossible to know where these bind without solving the structures. Using AlphaFold2 (AF) and other AI methods to distinguish true binders has proven highly successful but relies on the availability of binding scaffolds. Here, we develop EvoBind, an in silico directed-evolution platform based on AF that designs peptide binders towards an interface using only sequence information. We show that AF can distinguish between native and mutated peptide binders using the plDDT score and find that AF adapts the receptor interface structure to the binders during optimisation. We analyse previously designed minibinder proteins and show that AF can distinguish designed binders from non-binders. We compare ELISA ratios of different peptide binders and find the affinity can not be distinguished among binders, possibly due to varying binding sites and low AF confidence. We test the recovery of binding motifs and find that up to 75% of motifs are recovered. In principle, EvoBind can be used to design binders towards any interface conditioned on if AF can predict these. We expect that EvoBind will aid experimentalists substantially, providing a starting point for further laboratory analysis and optimisation. We hope that the use of AI-based methods will come to make binder design significantly cheaper and more accurate in tackling unmet clinical needs. EvoBind is freely available at: https://colab.research.google.com/github/patrickbryant1/EvoBind/blob/master/EvoBind.ipynb

bioinformatics↗

bioRxiv2022

Predicting the structure of large protein complexes using AlphaFold and sequential assembly

AlphaFold can predict the structure of single- and multiple-chain proteins with very high accuracy. However, the accuracy decreases with the number of chains, and the available GPU memory limits the size of protein complexes which can be predicted. Here we show that one can predict the structure of large complexes starting from predictions of subcomponents. We assemble 91 out of 175 complexes with 10-30 chains from predicted subcomponents using Monte Carlo tree search, with a median TM-score of 0.51. There are 30 highly accurate complexes (TM-score [≥]0.8, 33% of complete assemblies). We create a scoring function, mpDockQ, that can distinguish if assemblies are complete and predict their accuracy. We find that complexes containing symmetry are accurately assembled, while asymmetrical complexes remain challenging. The method is freely available and accesible as a Colab notebook https://colab.research.google.com/github/patrickbryant1/MoLPC/blob/master/MoLPC.ipynb.

bioinformatics↗

bioRxiv2021

Towards a structurally resolved human protein interaction network

All cellular functions are governed by complex molecular machines that assemble through protein-protein interactions. Their atomic details are critical to the study of their molecular mechanisms but fewer than 5% of hundreds of thousands of human interactions have been structurally characterized. Here, we test the potential and limitations of recent progress in deep-learning methods using AlphaFold2 to predict structures for 65,484 human interactions. We show that higher confidence models are enriched in interactions supported by affinity or structure-based methods and can be orthogonally confirmed by spatial constraints defined by cross-link data. We identify 3,137 high confidence models, of which 1,371 have no homology to a known structure, from which we identify interface residues harbouring disease mutations, suggesting potential mechanisms for pathogenic variants. We find groups of interface phosphorylation sites that show patterns of co-regulation across conditions, suggestive of coordinated tuning of multiple interactions as signalling responses. Finally, we provide examples of how the predicted binary complexes can be used to build larger assemblies. Accurate prediction of protein complexes promises to greatly expand our understanding of the atomic details of human cell biology in health and disease.

bioinformatics↗

bioRxiv2021

A structural biology community assessment of AlphaFold 2 applications

Most proteins fold into 3D structures that determine how they function and orchestrate the biological processes of the cell. Recent developments in computational methods have led to protein structure predictions that have reached the accuracy of experimentally determined models. While this has been independently verified, the implementation of these methods across structural biology applications remains to be tested. Here, we evaluate the use of AlphaFold 2 (AF2) predictions in the study of characteristic structural elements; the impact of missense variants; function and ligand binding site predictions; modelling of interactions; and modelling of experimental structural data. For 11 proteomes, an average of 25% additional residues can be confidently modelled when compared to homology modelling, identifying structural features rarely seen in the PDB. AF2-based predictions of protein disorder and protein complexes surpass state-of-the-art tools and AF2 models can be used across diverse applications equally well compared to experimentally determined structures, when the confidence metrics are critically considered. In summary, we find that these advances are likely to have a transformative impact in structural biology and broader life science research.

biophysics↗

bioRxiv2021

Improved prediction of protein-protein interactions using AlphaFold2 and extended multiple-sequence alignments

Predicting the structure of interacting protein chains is a fundamental step towards understanding protein function. Unfortunately, no computational method can produce accurate structures of protein complexes. AlphaFold2, has shown unprecedented levels of accuracy in modelling single chain protein structures. Here, we apply AlphaFold2 for the prediction of heterodimeric protein complexes. We find that the AlphaFold2 protocol together with optimized multiple sequence alignments, generate models with acceptable quality (DockQ[≥]0.23) for 63% of the dimers. From the predicted interfaces we create a simple function to predict the DockQ score which distinguishes acceptable from incorrect models as well as interacting from non-interacting proteins with state-of-art accuracy. We find that, using the predicted DockQ scores, we can identify 51% of all interacting pairs at 1% FPR. The protocol can be found at: https://gitlab.com/ElofssonLab/FoldDock.

bioinformatics↗

bioRxiv2021

The relationship between ageing and changes in the human blood and brain methylomes

Changes in DNA methylation have been found to be strongly correlated with age, enabling the creation of "epigenetic clocks". Previously, studies on the relationship between ageing and DNA methylation have assumed a linear relationship, but no study has shown this always to be the case. We show that the relationships between significant methylation changes and ageing are different in different tissues, and that these changes show variable rates across life. Further, we observe a tendency for saturation as ageing proceeds. We provide a straightforward method of assessing all methylation-age relationships and cluster them according to their relative change rates based on fold change. Less than 13% of the significant markers selected here are used in the most common epigenetic clocks. When the significant markers display largely non-linear relationships, our fold change selection outperforms the most common epigenetic clocks in predicting age. We also show that the saturation observed at older ages explains the earlier observations that the epigenetic clock consistently underestimates the age in older samples. The findings imply that markers selected from linear adjustments or correlations do not represent the most meaningful biological changes.

bioinformatics↗