Biology subjects

Jenkins, T. P.

Publications and source records attributed to Jenkins, T. P..

At least 19 recordsLinked to original sources

bioRxiv2026

A rapid, automation-friendly, cell-free and purification-free pipeline for binding and kinetic characterisation of de novo designed binders with high quality standardised per-design data

De novo protein binders can now be generated far faster than they can be tested. Current validation routes - display selection, cell-free two-hybrid assays, or surface-based biophysics - require cell culture, colony picking, protein purification, target labeling or sensor immobilisation, take 2 to 8 weeks, and require additional methods for affinity and kinetics characterisation. Here we report an end-to-end, cell- and purification-free pipeline that goes from gene to binding affinity and kinetics in three days without genetically modified organisms, cell culture, purification, labelling or immobilisation. An automated, metric-driven loop of RFdiffusion, ProteinMPNN and AlphaFold2 initial-guess generates and filters designs, which are ranked by Boltz-2 ipSAE, assembled by Golden Gate cloning from linear DNA, expressed in a cell-free protein synthesis lysate as fluorescent protein fusions, and screened directly in the unpurified product by flow-induced dispersion analysis (FIDA). Against human growth hormone, 11 of 30 designs (37%) bound, and the four strongest had equilibrium dissociation constants of 0.93-9.98 nM with on- and off-rates resolved from the same cell-free material. Each sample run measures expression level, hydrodynamic radius, oligomeric and aggregation state alongside the binding readout, producing the standardised, per-design datasets that model training requires.

molecular biology↗

bioRxiv2026

Specificity-driven protein binder design with Odin-Multi

A useful protein binder is defined as much by what it does not bind as by what it does. Some applications call for one binder to cover a family of related targets; others require it to distinguish a single member from near-identical relatives. Yet, widely used deep-learning-based de novo design methods typically optimise one interaction at a time, leaving cross-reactivity and specificity to emerge during downstream screening. Here we present Odin-Multi, a binder design framework that optimises a shared binder sequence against several complexes simultaneously, applying attractive objectives to on-targets and repulsive objectives to off-targets. We benchmarked Odin-Multi in silico across three systems representing distinct cross-reactivity and specificity challenges: class B1 G protein-coupled receptors (GPCRs), testing cross-reactivity across multiple therapeutically relevant receptors; short-chain three-finger toxins, testing cross-reactivity across homologous toxin family members; and peptide-MHC (pMHC) complexes, testing specificity between near-identical target and off-target surfaces. For pairs of related class B1 GPCRs, 83.5 to 96.8% of jointly optimised designs exceeded an interaction-confidence threshold for both targets, compared with 6.8 to 36.3% of designs from single-target campaigns. For two short-chain three-finger neurotoxins, 9.2% of jointly optimised designs exceeded the corresponding threshold for both targets, compared with 0.8% of designs optimised against one toxin alone. Finally, in a pMHC specificity benchmark where target and off-target differed only in a single peptide residue, counter-selection increased the fraction of designs satisfying both the target-confidence criterion and a target-to-off-target interaction-confidence ratio of 2.5 from 6.0% to 14.2%. Experimental screening produced leads consistent with both design regimes in the two systems tested in vitro. We identified a cross-reactive toxin minibinder showing apparent nanomolar binding to the neurotoxin Erabutoxin A and to a candidate NK-shNTx-containing fraction from Naja kaouthia venom (higher-affinity fitted components of 11.95 and 34.43 nM, respectively), and a pMHC minibinder with greater target-to-off-target discrimination than a previously reported design. By treating cross-reactivity and specificity as explicit design objectives rather than screening outcomes, Odin-Multi widens the range of binding behaviours accessible to computational design.

bioengineering↗

bioRxiv2026

Learning from tandem mass spectra at scale with a self-supervised foundation model for proteomics

Mass spectrometry-based proteomics increasingly relies on machine learning, yet existing models are trained for defined supervised tasks such as peptide identification, de novo sequencing or fragment intensity prediction, limiting transfer across datasets, instruments and acquisition methods. Here we present InstaNovo-FM, a self-supervised foundation model for bottom-up proteomics trained to reconstruct masked regions of tandem mass spectra. We assemble a diverse training corpus spanning 1.63 billion MS/MS spectra and 184.6 million high-confidence annotations. We train an encoder-only transformer on the annotated tier using a physics-aware masked reconstruction objective. We demonstrate that the InstaNovo-FM embeddings encode fundamental experimental and biological properties, including fragmentation method, sequence properties and post-translational modifications, without requiring peptide labels. Furthermore, this foundation model directly enables diverse downstream applications, including de novo peptide sequencing, database-free identification and analytical run classification. InstaNovo-FM establishes a unified representation space for peptide fragmentation spectra, enabling robust transferability across the proteomics ecosystem.

bioinformatics↗

bioRxiv2026

De novo designed single-domain antibodies protect against lethal cobra venom neurotoxicity in vivo

Generative protein design can now rapidly produce de novo binders with high affinity and functional activity against a wide range of targets, including lethal snake venom toxins. However, so far most reported successes rely on new-to-nature scaffolds with limited therapeutic precedent. Single-domain antibodies (VHHs) offer a clinically validated alternative scaffold that can bind and neutralize long-chain -neurotoxins, which are some of the most lethal components in snake venoms. Here we compare three recently established de novo design models with VHH-design capabilities (Germinal, RFantibody, and BoltzGen) for their ability to generate VHHs against the neurotoxin -cobratoxin from the monocled cobra (Naja kaouthia). Using standardized model inputs and evaluation criteria based on AlphaFold3 interface confidence (ipTM) and RMSD self-consistency, we find that Germinal was the only method to generate designs passing stringent in silico criteria for experimental testing. We therefore performed a larger Germinal design campaign employing three different VHH frameworks and experimentally validated 46 designs in vitro. Of these, 42 expressed as soluble proteins and we identified four binding hits derived from two of the three tested frameworks. Of the four binders, two lead candidates were further characterized and demonstrated high affinity (KDs of 4.1 nM and 10.8 nM), monomeric behavior and low polyreactivity, indicating favorable biophysical and developability properties, as well as functional toxin neutralization in vitro. To assess their therapeutic potential we investigated their ability to protect against -cobratoxin toxicity in vivo. Both candidates fully protected mice after -cobratoxin challenge, with 100% survival compared to a lethal control. One candidate also retained notable neutralization capacity against whole venom of Naja kaouthia with a survival of 56%, while the other protected 22% when tested in a rescue setting. Together, we demonstrate that de novo VHH design can generate high affinity single-domain antibodies with in vivo protection against lethal cobra venom neurotoxicity, and provide practical insights into method- and framework-dependent performance.

bioengineering↗

bioRxiv2026

De novo designed cyclic MC4R peptide agonist reduces food intake in mice

Deep learning-based structure prediction enables the design of peptide ligands without relying on naturally occurring scaffolds. However, most computationally generated peptides are not advanced beyond initial activity measurements, leaving the path to drug-like optimization and in vivo validation underexplored. Here we establish an end-to-end workflow for de novo peptide agonist discovery and maturation using the melanocortin-4 receptor (MC4R) as a model target. Using an AlphaFold2-based hallucination protocol implemented in ColabDesign, we generated more than 5,000 linear and head-to-tail cyclic candidate peptides directed towards the MC4R orthosteric pocket. Functional screening of a prioritized subset revealed measurable activity in 74% of linear peptides and 23% of cyclic peptides, from which we identified a cyclic agonist with an EC50 of 340 nM despite lacking the canonical melanocortin activation motif. We then performed systematic in vitro maturation by deep mutational scanning, half-life extender conjugation scanning, and a combinatorial optimization library, coupled with data-driven analysis to map sequence-activity relationships. These experiments identified an alternative activation motif centered on an APWR segment and yielded single-site variants with substantially improved potency. The most effective substitution, a proline at position 5, produced the E5P variant with an EC50 of 6.7 nM against the human melanocortin-4 receptor (hMC4R). Finally, central administration of E5P (10 nmol) reduced acute food intake in mice, providing in vivo proof of concept. Together, our results demonstrate a generalizable design-to-validation strategy for converting de novo peptide designs into optimized, pharmacologically active peptides, and expand the space of MC4R agonist chemotypes beyond endogenous melanocortins.

molecular biology↗

bioRxiv2026

Accelerating multi-objective VHH discovery via integrated high-throughput selection and AlphaFold3-guided structure prediction

Discovering therapeutic antibodies that bind multiple related targets with high affinity and favourable biophysical properties remains challenging and resource intensive. For snakebite antivenoms, this challenge is critical as treatments must neutralise toxins across multiple snake species. We developed a pipeline combining high-throughput yeast screening, deep sequencing, and AlphaFold3 structure prediction to rapidly identify poly-specific variable domains of heavy-chain-only antibodies (VHHs) against long-chain -neurotoxins. Multiplexed yeast display screening generated a dataset of diverse candidates with varying binding specificities. AlphaFold3-generated VHH-toxin complex predictions enabled structure-based prioritisation that accurately predicted poly-specific binders targeting conserved epitopes across multiple toxins. These structural insights enabled computational optimisation of both affinity and solubility without disrupting target recognition. Experimental validation confirmed improved variants maintained broad specificity across toxins. This integrated approach accelerates multi-objective antibody discovery by predicting which candidates will bind multiple targets before extensive laboratory testing, providing a generalisable strategy applicable beyond antivenoms to any therapeutic requiring broad target coverage.

bioengineering↗

bioRxiv2025

Clearing the whey: Efficient inhibition and removal of chymosin from whey by de novo designed protein-based inhibitors

Chymosins are indispensable as coagulants in cheese production. However, the residual activity of these enzymes in the whey by-product limits downstream processing for further food applications, e.g. infant milk formulations or protein powder for athletic purposes. To date, inactivation of coagulants and other enzymes is primarily achieved through heat inactivation (pasteurisation), which remains a challenging and energy-intensive task at an industrial scale and may have adverse effects on the quality. Here, we report the de novo design of small protein minibinders (miBds) that inhibit coagulant activity which marks one of the first industrial applications of generative binder design. Using RFdiffusion1, ProteinMPNN2, and AlphaFold23, we generated and filtered over 2,000 backbone designs to a final panel of 63 candidates targeting bovine and camel chymosins. The first round of selected binders exhibited low-nanomolar Kiapp 4,5 values against chymosins originating from bovine, camel, and hedgehog, with reversible, pH-dependent dissociation profiles. Notably, we identified miBds specific to bovine and camel chymosin variants, as well as broadly cross-reactive including coagulants of fungal origin. Moreover, when immobilised, the miBds efficiently removed residual chymosin from whey. It was possible to recycle the miBd-coated resin repeatedly through treatment of the resin with high pH (> 11) or low pH (< 3) which emulates industry standard clean-in-place procedures. This dual functionality, of inhibition and reversible capture of chymosin, enables a versatile application for sustainable food production and precise control of proteases and can be extended to other enzyme classes in dairy.

bioengineering↗

bioRxiv2025

De novo designed protein enables precise epitope-level control of Gremlin-1 antagonism

Growth-factor antagonists regulate key developmental and pathological processes, yet the molecular principles governing their control remain poorly understood. Gremlin-1 (GREM1) is a secreted antagonist of bone morphogenetic proteins (BMPs) whose dysregulation is implicated in fibrosis and cancer. Here, we report a de novo designed protein that binds GREM1 with sub-nanomolar affinity and selectively releases BMPs for downstream signalling. By integrating generative deep-learning-based protein design with molecular-dynamics-derived flexibility descriptors, we identify a predictive relationship between interface rigidity, desolvation energy, and binding success. The resulting binder, RF1-2, reproduces the native BMP-binding epitope on GREM1 at near-atomic precision, as confirmed by cryo-electron microscopy, and competitively blocks BMP-2 and BMP-4 association. These results establish interface rigidity as a key physical determinant of antagonist inhibition and demonstrate how AI-guided protein design can uncover molecular principles underlying extracellular signalling control.

biochemistry↗

bioRxiv2025

Predicting Experimental Success in De Novo Binder Design: A Meta-Analysis of 3,766 Experimentally Characterised Binders

Designing high-affinity de novo protein binders has become increasingly tractable, yet in vitro prioritisation continues to depend on heuristics in the absence of systematic analysis. Here, we present a large-scale meta-analysis of 3,766 experimentally tested binders across 15 structurally diverse targets. Using a unified, high-throughput pipeline that re-predicts each binder-target complex with AF2 (initial guess and ColabFold), AF3 and Boltz-1, we extract over 200 structural, energetic and confidence features per design. We show that interface-focused metrics, most notably the AF3-derived interaction prediction Score from Aligned Errors (ipSAE) outperform commonly used scores such as ipAE and ipTM, with a significant 1.4-fold increase in average precision compared to ipAE. We further show that combining these metrics with orthogonal physicochemical interface descriptors, including Rosetta {Delta}G/{Delta}SASA and interface shape complementarity, improves predictive performance. While overall per-formance varies by target, simple linear models trained on a small number of AF3-derived features generalize well across datasets. We propose interpretable, target-agnostic filtering strategies, such as combining AF3 ipSAE_min rankings with structural filters, to improve precision in selecting binders for testing. Finally, we release the complete dataset establishing a community resource to benchmark and accelerate de novo binder discovery.

bioinformatics↗

bioRxiv2025

Generalizable direct protein sequencing with InstaNexus

Protein-based therapeutics, such as antibodies and nanobodies, are not encoded in reference genomes, challenging their accurate characterization via standard proteomics. Current methods rely on indirect inference, fragmented outputs, and labor-intensive workflows, which hinder functional insights and routine application. Here, we present a generalizable, end-to-end workflow for direct protein sequencing, combining streamlined sample preparation, AI-driven de novo peptide sequencing, and tailored assembly to reconstruct contiguous protein sequences. A novel composite scoring framework prioritises longer assemblies and coverage, enhancing accuracy and reducing ambiguity. Validation across diverse protein modalities demonstrates its utility and ability to robustly sequence functionally critical regions of selected proteins. This workflow represents an advance in precision proteomics with promising applications in therapeutic discovery, immune profiling, and protein science.

bioinformatics↗

bioRxiv2025

InstaNovo-P: A de novo peptide sequencing model for phosphoproteomics

Phosphorylation, a crucial post-translational modification (PTM), plays a central role in cellular signaling and disease mechanisms. Mass spectrometry-based phosphoproteomics is widely used for system-wide characterization of phosphorylation events. However, traditional methods struggle with accurate phosphorylated site localization, complex search spaces, and detecting sequences outside the reference database. Advances in de novo peptide sequencing offer opportunities to address these limitations, but have yet to become integrated and adapted for phosphoproteomics datasets. Here, we present InstaNovo-P, a phosphorylation specific version of our transformer-based InstaNovo model, fine-tuned on extensive phosphoproteomics datasets. InstaNovo-P significantly surpasses existing methods in phosphorylated peptide detection and phosphorylated site localization accuracy across multiple datasets, including complex experimental scenarios. Our model robustly identifies peptides with single and multiple phosphorylated sites, effectively localizing phosphorylation events on serine, threonine, and tyrosine residues. We experimentally validate our model predictions by studying FGFR2 signaling, further demonstrating that InstaNovo-P uncovers phosphorylated sites previously missed by traditional database searches. These predictions align with critical biological processes, confirming the models capacity to yield valuable biological insights. InstaNovo-P adds value to phosphoproteomics experiments by effectively identifying biologically relevant phosphorylation events without prior information, providing a powerful analytical tool for the dissection of signaling pathways.

bioinformatics↗

bioRxiv2024

De novo designed pMHC binders facilitate T cell induced killing of cancer cells

The recognition of intracellular antigens by CD8+ T cells through T-cell receptors (TCRs) is central to adaptive immunity, enabling responses against infections and cancer. The recent approval of TCR-gene-edited T cells for cancer therapy demonstrates the therapeutic advantage of using pMHC recognition to eliminate cancer. However, identification and selection of TCRs from patient material is complex and influenced by the TCR repertoire of the donors used. To overcome these limitations, we here present a rapid and robust de novo binder design platform leveraging state-of-the-art generative models, including RFdiffusion, ProteinMPNN, and AlphaFold2, to engineer minibinders (miBds) targeting the cancer-associated pMHC complex, NY-ESO-1(157-165)/HLA-A*02:01. By incorporating in silico cross-panning and molecular dynamics simulations, we enhanced specificity screening to minimise off-target interactions. We identified a miBd that exhibited high specificity for the NY-ESO-1-derived peptide SLLMWITQC in complex with HLA-A*02:01 and minimal cross-reactivity in mammalian display assays. We further demonstrate the therapeutic potential of this miBd by integrating it into a chimeric antigen receptor, as de novo Binders for Immune-mediated Killing Engagers (BIKEs). BIKE-transduced T cells selectively and effectively killed NY-ESO-1+ melanoma cells compared to non-transduced controls, demonstrating the promise of this approach in precision cancer immunotherapy. Our findings underscore the transformative potential of generative protein design for accelerating the discovery of high-specificity pMHC-targeting therapeutics. Beyond CAR-T applications, our workflow establishes a foundation for developing miBds as versatile tools, heralding a new era of precision immunotherapy.

immunology↗

bioRxiv2024

Systematic benchmarking of mass spectrometry-based antibody sequencing reveals methodological biases

The circulating antibody repertoire is crucial for immune protection, holding significant immunological and biotechnological value. While bottom-up mass spectrometry (MS) is the most widely used proteomics technique for profiling the sequence diversity of circulating antibodies (Ab-seq), it has not been thoroughly benchmarked. We quantified the replicability and robustness of Ab-seq using six monoclonal antibodies with known protein sequences in 70 different combinations of concentration and oligoclonality, both with and without polyclonal serum IgG background. Each combination underwent four protease treatments and was analyzed across four experimental and three technical replicates, totaling 3,360 LC-MS/MS runs. We quantified the dependence of MS-based Ab-seq identification on antibody sequence, concentration, protease, background signal diversity, and bioinformatics setups. Integrating the data from experimental replicates, proteases, and bioinformatics tools enhanced antibody identification. De novo peptide sequencing showed similar performance to database-dependent methods for higher antibody concentrations, but de novo antibody reconstruction remains challenging. Our work provides a foundational resource for the field of MS-based antibody profiling.

immunology↗

bioRxiv2024

Improved protein binder design using -pairing targeted RFdiffusion

Despite recent advances in the computational design of protein binders, designing proteins that bind with high affinity to polar protein targets remains an outstanding problem. Here we show that RFdiffusion can be conditioned to efficiently generate protein scaffolds that form geometrically matched extended beta-sheets with target protein edge beta-strands in which polar groups on the target are nearly perfectly complemented with hydrogen bonding groups on the design. We use this approach to design binders against a set of therapeutically relevant polar targets (KIT, PDGFR[a], ALK-2, ALK-3, FCRL5, and NRP1) and find that beta-strand-targeted design yields higher affinities and success rates than unconditioned RFdiffusion. All by all binding experiments show that the designs have affinities ranging from 137 pM to mid nM for their targets and essentially no off target binding despite the sharing of beta-strand interactions, likely reflecting the precise customization of interacting beta-strand geometry and additional designed binder-target interactions. A co-crystal structure of one such design in complex with the KIT receptor is nearly identical to the computational design model confirming the accuracy of the design approach. The ability to robustly generate binders displaying high affinity and specificity to polar interaction surfaces with exposed beta-strands considerably increases the range and capabilities of computational binder design.

biochemistry↗

bioRxiv2024

ProteusAI: An Open-Source and User-Friendly Platform for Machine Learning-Guided Protein Design and Engineering

AO_SCPLOWBSTRACTC_SCPLOWProtein design and engineering are crucial for advancements in biotechnology, medicine, and sustainability. Machine learning (ML) models are used to design or enhance protein properties such as stability, catalytic activity, and selectivity. However, many existing ML tools require specialized expertise or lack open-source availability, limiting broader use and further development. To address this, we developed ProteusAI, a user-friendly and open-source ML platform to streamline protein engineering and design tasks. ProteusAI offers modules to support researchers in various stages of the design-build-test-learn (DBTL) cycle, including protein discovery, structure-based design, zero-shot predictions, and ML-guided directed evolution (MLDE). Our benchmarking results demonstrate ProteusAIs efficiency in improving proteins and enyzmes within a few DBTL-cycle iterations. ProteusAI democratizes access to ML-guided protein engineering and is freely available for academic and commercial use. Future work aims to expand and integrate novel methods in computational protein and enzyme design to further develop ProteusAI.

bioinformatics↗

bioRxiv2024

Global diversity of soil-transmitted helminths reveals population-biased genetic variation that impacts diagnostic targets

Soil-transmitted helminths (STHs) are intestinal pathogens affecting over a billion people worldwide. STH control relies on microsopy-based diagnostics to monitor parasite prevalence and enable post-treatment surveillance, however, molecular diagnostics are rapidly being developed due to increased sensitivity, particularly in low-STH-prevalence settings. The genetic diversity of helminths and its potential impact on molecular diagnostics remains unclear. Using low-coverage genome sequencing, we assessed the presence of STHs in worm, faecal and purified egg samples from 27 countries, identifying differences in the genetic connectivity and diversity of STH-positive samples across regions and cryptic diversity between closely related human- and pig-infective species. We defined substancial copy number and sequence variants in current diagnostic targets and validated the impact of genetic variants on qPCR diagnostics using in vitro assays. Our study provides insights into the diversity and genomic epidemiology of STHs, highlighting challenges and opportunities for developing molecular diagnostics needed to support STH control efforts.

genomics↗

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

mRNA-delivered consensus allergens induce a neutralizing IgG response against food and pollen allergens

Pollen-food allergy syndrome (PFAS) affects a significant proportion of the global population with a major health and socioeconomic impact. Patients are generally treated against the major sensitized allergen which does not warrant protection against cross-reactive allergens, leading to long and ineffective treatment regimens. For food allergies, patient guidelines rely on source avoidance, leading to dietary restrictions and reduced quality of life - in particular for those suffering from PFAS. To overcome these limitations, we introduce a novel allergy immunotherapy (AIT) approach utilizing consensus allergens and mRNA technology to achieve broader, safer, and faster desensitization in PFAS patients. We first designed a consensus allergen of orthologs of non-specific Lipid Transfer Proteins (cnsLTP-1) representing a broad spectrum of nsLTP allergens prevalent in food and pollen sources. CnsLTP-1 was delivered to naive BALB/c mice using mRNA-lipid nanoparticles (mRNA-LNP) as vehicle, or by a traditional protein formulation, to assess if it elicits broad protection against allergens from different sources. Immunization with both mRNA-LNP and protein formulations demonstrated that cnsLTP-1-specific IgGs could be induced, whilst the mRNA-LNP formulation notably avoided the induction of allergen-specific IgEs. The induced antibodies were capable of recognizing and binding to a variety of nsLTPs, and effectively blocked the binding of allergens by allergic patient serum IgEs. This study thus demonstrates that the presented AIT strategy, based on mRNA-LNP technology and consensus allergens, could find clinical utility by addressing the limitations of current AIT. Further development of this technology platform could pave the way for more effective and patient-friendly treatments for PFAS and other cross-reactive allergies.

bioengineering↗