Biology subjects

Nielsen, M.

Publications and source records attributed to Nielsen, M..

At least 19 recordsLinked to original sources

bioRxiv2026

Exploring noise in TCR specificity data through supervised learning

Accurate prediction of T cell receptor (TCR) specificity can potentially accelerate development of novel immunotherapies, yet public databases used to train these models contain substantial label noise. This uncertainty complicates model evaluation and limits the reliability of predicted TCR-pMHC interactions. In this work, we present a simple supervised machine learning denoising scheme to identify low-quality labels in TCR specificity data. We evaluated the framework using the TCRvdb validation dataset and further corroborated it via structural modeling with AlphaFold 3. Here, our denoising approach successfully identified non-validated TCRs, achieving an average MCC of 0.47 for filtering non-validated examples while retaining high sensitivity for validated examples. Applying our denoising approach to the complete NetTCR training dataset, we observed that data associated with known noisy attributes were significantly enriched in the filtered subset. Furthermore, TCR-pMHC complexes identified as noise consistently produced lower AlphaFold 3 confidence scores, providing independent evidence of misannotation. Experimental validation of 30 selected TCRs further supported the approach, with 80% of TCRs predicted as noise failing to demonstrate functional reactivity. Our findings demonstrate that current performance metrics may be obscured by dataset inaccuracies and provide a method for refining training data in the absence of gold-standard benchmarks.

bioinformatics↗

bioRxiv2026

Bispecific GD2xB7-H3 Antibody Improves Tumor Targeting and Reduces Toxicity while Maintaining Efficacy for Neuroblastoma

The current treatment regimen for neuroblastoma involves immunotherapy, including a monoclonal antibody that recognizes disialoganglioside (GD2), expressed at high levels on neuroblastoma. GD2 is not present on most normal tissues except for nerves. Thus, anti-GD2 antibody treatment causes substantial, dose-limiting neuropathic pain. B7-H3 is overexpressed on multiple tumor types, including neuroblastoma, and is largely absent on nerves and other normal tissues. We designed a bispecific antibody (bsAb) that requires simultaneous binding of these two tumor antigens to achieve tight binding of tumor cells. Our preclinical research shows that, compared with a monospecific anti-GD2 antibody, the GD2xB7-H3 bsAb has improved tumor specificity, comparable antitumor efficacy, and reduced nerve binding and pain-associated toxicity. Since this bsAb does not bind to nerves, it may permit more tolerable and sustained treatment schedules than are currently feasible with monospecific anti-GD2 antibodies. In addition, its enhanced tumor specificity may support future development as a targeted delivery platform for antibody-drug conjugates or other payload-based therapies, potentially improving both efficacy and quality of life for patients with neuroblastoma.

cancer biology↗

bioRxiv2026

TCRdenoise - an unsupervised similarity-based approach for denoising of TCR-pMHC specificity data

Public repositories of T cell receptor (TCR)-peptide-MHC (pMHC) interactions constitute a critical resource for studying adaptive immunity and developing predictive models of TCR specificity. However, recent evidence suggests that a substantial fraction of reported TCR-pMHC interactions may be incorrectly annotated, limiting the quality of downstream analyses and machine learning applications. Here, we present an unsupervised sequence similarity-based framework for denoising peptide-specific TCR repertoires. The method combines pairwise TCR similarity metrics derived from TCRbase and TCRdist3 with hierarchical clustering and a novel adaptation of the silhouette score designed to address the prevalence of singleton clusters and highly imbalanced cluster structures. By incorporating a pseudo-cluster containing singleton and background TCRs, and by optimising both clustering distance thresholds and minimum cluster-size criteria, the proposed approach identifies TCRs likely to represent true antigen-specific binders while filtering putative noise. Using experimentally validated repertoires from TCRvdb, we demonstrate that the modified silhouette score closely tracks clustering solutions that maximise separation between binding and non-binding TCRs, achieving strong agreement with independent validation based on the Matthews correlation coefficient. Extension to a large collection of peptide-specific TCR data revealed a strong negative correlation between the percentage of TCRs classified as noise and the predictive performance of peptide-specific binding models. Further, the denoising classification labels on this data set were corroborated using structural modeling confidence scores of the peptide-TCR interface extracted from a refined AlphaFold 3 modeling pipeline. Additionally, retraining NetTCR on denoised data improved internal cross-validated performance compared with models trained on the full data set, whereas models trained exclusively on TCRs classified as noise performed close to random. Together, these results demonstrate that sequence similarity-based denoising can effectively enrich for biologically meaningful TCR-pMHC interactions and improve the quality of training data for predictive immunological models. The proposed framework provides a scalable strategy for improving the reliability of public TCR databases and facilitating the development of more accurate TCR specificity prediction methods.

bioinformatics↗

bioRxiv2026

Label Noise Limits TCR-pMHC Specificity Prediction: Improved Performance Through AlphaFold3-Based Structural Modeling and Data Denoising

T cell receptor (TCR) binding to peptides presented by major histocompatibility complex (MHC) molecules is a key step in T cell activation, and forms the basis of adaptive immunity. Predicting this specificity is therefore essential to developing effective TCR-based immunotherapies and vaccines. Despite its clinical relevance, predicting TCR-pMHC specificity for previously unseen peptides remains an open problem, with structural modeling so far the only strategy showing any predictive power in this setting. In this study, we find that this limited performance is substantially driven by label noise in the data used to train and evaluate these methods, an effect that has so far been largely underexplored. Using an AlphaFold3-based pipeline adapted for TCR-pMHC structural modeling, we achieve state-of-the-art specificity prediction, outperforming AlphaFold2.3-based and sequence based methods, and performing at par with the leading Immrep2025 competition submission. Combining this pipeline with a cluster-based denoising algorithm, we show that removing mislabeled points from a large specificity dataset increased binder ranking accuracy by more than 70% relative to the full dataset. Together, these results highlight label noise as a major factor limiting the performance that any method in this field can achieve, and show that combining structural modeling with label denoising substantially improves TCR-pMHC specificity prediction, making such approaches an attractive complement to current sequence-based approaches for refining TCR target selection.

bioinformatics↗

bioRxiv2026

Beyond natural amino acids: Extending immunogenicity risk assessment to non-canonical peptide drugs through chemical feature encoding

Peptide therapeutics are increasingly used to treat challenging diseases, but immunogenicity risks limit their clinical success. In silico tools enable immunogenicity screening through prediction of peptide-MHCII binding, yet current methods fail to capture chemical properties of non-natural amino acids routinely incorporated to improve drug properties. Here, we present a machine learning approach combining chemical fingerprints with sequence information to predict MHC class II binding for both canonical and modified peptides. We propose two molecular representations (direct-encoding and similarity-based chemical fingerprints) that preserve positional information while encoding chemical diversity. These representations achieved performance comparable to sequence-based encodings (BLOSUM62 and one-hot) for canonical peptides while accurately identifying binding cores and motifs. Testing on citrullinated peptides, chemical fingerprints substantially improved quantitative prediction accuracy while maintaining comparable linear correlation across encoding methods, demonstrating the importance of explicit chemical representation for accurate absolute binding affinity prediction. These descriptors can be integrated into pan-allele prediction frameworks, enabling immunogenicity risk assessment across diverse modifications and therapeutic modalities, including peptide therapeutics, antibody-drug conjugates, and synthetic vaccines. The proposed chemistry-informed framework addresses a critical gap in preclinical drug development, facilitating early mitigation strategies before costly clinical trials.

bioinformatics↗

bioRxiv2026

IMMREP25: Unseen Peptides

T cell receptors (TCRs) can bind to peptides presented by MHC molecules (pMHC) as a first step to trigger a T cell response. Reliable approaches to predict TCR:pMHC binding would have broad applications in clinical diagnostics, therapeutics, and the fundamental understanding of molecular interactions. IMMREP is a community organized series of prediction contests that asks participants to predict TCR:pMHC binding on unpublished datasets. Previous iterations in 2022 and 2023 showed multiple approaches can predict TCR-pMHC binding with significant accuracy (median AUC_0.1[≥]0.7) for peptides where experimental data is available ("seen" peptides). In contrast, models did not outperform random guessing for peptides that have no such data available ("unseen" peptides). Here we report on the results of IMMREP25, which focused solely on unseen peptides in order to evaluate the cutting edge of the field. We received 126 named submissions predicting the specificity of 1,000 TCRs against twenty unseen peptides restricted by one of two MHC molecules (HLA-A*02:01 and HLA-B*40:01). The best performing methods showed a macro-AUC_0.1 of 0.60, significantly better than random, demonstrating significant advances in the field. The top performing methods incorporated structural modeling into their approach, indicating that especially for unseen peptides, a structural understanding aids in the prediction of TCR:pMHC interactions. The results from this benchmark highlight the significant challenges remaining for TCR:pMHC predictions and will inform future method development.

bioinformatics↗

bioRxiv2026

NetMHCIIphosPan: a machine learning tool for predictingHLA class II antigen presentation of phosphorylated peptides.

Phosphorylated peptides presented by human leukocyte antigen (HLA) class II molecules play pivotal roles in immune regulation, yet their characterization and prediction remain challenging due to data noise and limited HLA coverage. Here, we introduce NetMHCIIphosPan, a prediction method for HLA-II antigen presentation of phosphorylated peptides, developed using mass spectrometry (MS)-based immunopeptidomics datasets. Employing a refined peptide identification workflow, we reanalyzed earlier HLA-II phospholigand datasets and trained predictive models, achieving superior performance compared to models trained on original data. Binding motif analysis revealed that HLA-specific preferences for phospholigands closely aligned with those of unmodified ligands. Incorporating unmodified ligands into training further enhanced predictive accuracy, particularly for HLA-DP and HLA-DQ molecules. NetMHCIIphosPan outperformed existing tools, such as NetMHCIIpan-4.3 and MixMHC2pred-1.3, for prediction of HLA antigen presentation of phosphorylated peptides demonstrating robustness and utility. This work establishes NetMHCIIphosPan as a state-of-the-art tool for understanding the HLA-II phospholigandome, with potential applications in immunotherapy and vaccine design.

bioinformatics↗

bioRxiv2025

ITRAP2, a flexible and robust strategy to assign antigen recognition of T-cells in a coupled single-cell TCR-pMHC assay

Determining T-cell specificity forms a crucial step toward understanding T-cell involvement in health and disease. Single-cell sequencing technologies allow for co-capture of TCR alpha and beta chains, and their antigen specificity can be determined through peptide-MHC (pMHC) multimer binding and capture of a co-attached barcode oligo. However, SC sequencing often includes a high level of dropouts and risk of cross-contamination. Similarly, barcoded pMHC readouts often suffer from significant background noise. These issues complicate the automatic assignment of pMHC recognition to TCR clonotypes. To overcome these challenges, we developed a method for data denoising - Improved T-cell Receptor Antigen Paring 2 (ITRAP2). This approach significantly reduces noise in single-cell pMHC readouts and allows for accurate identification of TCR specificity. ITRAP2 incorporates statistical tests and confidence metrics for each TCR-pMHC pairing, offering user flexibility in pairing rigor, and allows multiple pMHC assignments to the same T-cell clone in the event of cross-binding within the pMHC multimer library. We tested this method on an in-house generated dataset of 8141 single cells, screened for CD8 T-cell binding using a panel of 100 different barcode-labelled pMHC multimers holding virus-derived peptides, and on a larger public dataset from 10x Genomics with 208,589 T-cells evaluated for recognition using a panel of 50 different pMHCs. In both datasets, ITRAP2 was able to recover TCR-pMHC hits that were missed either when investigating the raw data or analyzing the data using alternative tools. Importantly, we demonstrate that the size of the pMHC multimer library is crucial for accurate pMHC-TCR pairing and that a minimum of 25 pMHC multimer should be included to optimally determine background characteristic, and assigning true positive events.

bioinformatics↗

bioRxiv2025

Pocket Restraints Guided by B-Cell Epitope Prediction improves Chai-1 Antibody-Antigen Structure Modeling

The accurate prediction of antibody-antigen (AbAg) complexes is a key challenge for computational immunology, with applications in therapeutic antibody design and diagnostics. Current deep learning methods, such as AlphaFold and Chai, have the potential to generate high-confidence AbAg structures. However, these methods often fail to predict the correct AbAg structure, placing the antibody incorrectly on the antigen and converging on repeatedly predicting the same redundant binding mode. Here, we present BepiPocket and DiscoPocket, two simple approaches that integrate B-cell epitope prediction tools to guide antibody-epitope restraints during Chai-1 structure prediction. On a dataset of 1628 antibody-antigen complexes, we demonstrate that using the sequence-based predictor BepiPred-3.0 (BepiPocket) and the structure-based predictor DiscoTope-3.0 (DiscoPocket), we substantially improved both the accuracy and diversity of predicted AbAg complexes compared to standard Chai-1 modeling with random seed variation. We also find that a key driver in these performance gains is the antigen modeling accuracy. The software for both BepiPocket and DiscoPocket algorithms is freely available at https://github.com/mnielLab/BepiPocke

bioinformatics↗

bioRxiv2025

Characterizing antibodies binding to the same epitope reveals limited contribution by heavy chain CDR3 sequences relative to other CDRs

The antigen specificity of antibodies is defined mainly by their highly-variable complementarity-determining regions (CDRs), of which the CDR3 is considered to be the most important. Similar antibodies are expected to bind the same epitope on an antigen, but how this similarity should be quantified is unclear. Here, we utilized a large body of publicly available data to identify features that are most informative in predicting if two antibodies are likely to bind to the same epitope. We examined features such as pairwise sequence identities, CDR structural similarity, and geometrical similarity of antibody paratopes. We found that, for antibodies with high overall sequence similarity, the CDR sequence identity alone is sufficient to predict if a pair of antibodies binds to the same epitope. Strikingly, the sequence identity of CDR1 and CDR2 had as much or more predictive power than CDR3, which is commonly thought of as the main determinant of antibody specificity. In addition, for antibody pairs with lower overall sequence identity, including structural information led to significant improvements in predictive performance. Guided by these results, we developed BCRMatch (https://github.com/IEDB/BCRMatch) which uses an ensemble machine-learning approach to predict if pairs of antibodies target the same epitope based on antibody sequence alone. This can be applied to identify potential targets of an antibody of interest by comparing it to antibodies with known specificities.

bioinformatics↗

bioRxiv2025

NetTCR-struc, a structure driven approach for prediction of TCR-pMHC interactions

Accurate modeling of T cell receptor (TCR)-peptide-major histocompatibility complex (pMHC) interactions is critical for understanding immune recognition. In this study, we present advances in structural modeling of TCR-pMHC class I complexes focusing on improving docking quality scoring and structural model selection using graph neural networks (GNN). We find that AlphaFold-Multimers confidence score in certain cases correlates poorly with DockQ quality scores, leading to overestimation of model accuracy. Our proposed GNN solution achieves a 25% increase in Spearmans correlation between predicted quality and DockQ (from 0.681 to 0.855) and improves docking candidate ranking. Additionally, the GNN completely avoids selection of failed structures. Additionally, we assess the ability of our models to distinguish binding from non-binding TCR-pMHC interactions based on their predicted quality. Here, we demonstrate that our proposed model, particularly for high-quality structural models, is capable of discriminating between binding and non-binding complexes in a zero-shot setting. However, our findings also underlined that the structural pipeline struggled to generate sufficiently accurate TCR-pMHC models for reliable binding classification, highlighting the need for further improvements in modeling accuracy.

bioinformatics↗

bioRxiv2025

Contrasting genetic differentiation of urban and rural populations of two grassland lepidopterans across Europe

Urbanisation is transforming environments globally. The altered abiotic conditions and biotic interactions in urban habitats impose divergent selection pressures on urban versus rural populations, while genetic drift may also be significant in typically small urban populations. A key question in urban evolution concerns the origin and spread of urban genotypes. Examples exist of both single and multiple origins of urban genotypes, but these have proven difficult to generalize. Here, we address genetic differentiation among urban populations, among rural populations, and between urban and rural populations. We conducted an extensive population genomic double digest restriction-site associated DNA sequencing analysis of two non-model grassland lepidopterans, Coenonympha pamphilus and Chiasmia clathrata, across Europe. The genetic population structures of the study species were strikingly different: Co. pamphilus showed strong population differentiation, while this was almost absent in Ch. clathrata, which instead showed signs of high current and past gene flow among populations. Results of Co. pamphilus are consistent with multiple origins of urban populations, and multiple origins seem plausible also in Ch. clathrata. These results suggest that past and large-scale population dynamics need to be integrated into urban evolution research, because population history affects urban evolutionary dynamics.

evolutionary biology↗

bioRxiv2024

Molecular characterization of the archaic HLA-B*73:01 allele reveals presentation of a unique peptidome and skewed engagement by KIR2DL2

HLA class I alleles of archaic origin may have been retained in modern humans because they provide immunity against diseases to which archaic humans had evolved resistance. According to this model, archaic introgressed alleles were somehow distinct from those that evolved in African populations. Here we show that HLA-B*73:01, a rare allotype with putative archaic origins, has a relatively rare peptide binding motif with an unusually long-tailed peptide length distribution. We also find that HLA-B*73:01 combines a restricted and unique peptidome with high-cell surface expression, characteristics that make it well-suited to combat one or a number of closely-related pathogens. Furthermore, a crystal structure of HLA-B*73:01 in complex with KIR2DL2 highlights differences from previously solved structures with HLA-C molecules. These molecular characteristics distinguish HLA-B*73:01 from other HLA class I alleles previously investigated and may have provided early modern human migrants that inherited this allele with a selective advantage as they colonized Europe and Asia.

biochemistry↗

bioRxiv2024

Binning meets taxonomy: TaxVAMB improves metagenome binning using bi-modal variational autoencoder

A common procedure for studying the microbiome is binning the sequenced contigs into metagenome-assembled genomes. Currently, unsupervised and self-supervised deep learning based methods using co-abundance and sequence based motifs such as tetranucleotide frequencies are state-of-the-art for metagenome binning. Taxonomic labels derived from alignment based classification have not been widely used. Here, we propose TaxVAMB, a metagenome binning tool based on semi-supervised bi-modal variational autoencoders, combining tetranucleotide frequencies and contig co-abundances with contig annotations returned by any taxonomic classifier on any taxonomic rank. TaxVAMB outperforms all other binners on CAMI2 human microbiome datasets, returning on average 40% more near-complete assemblies than the next best binner. On real long-read datasets TaxVAMB recovers on average 13% more near-complete bins and 14% more species. When used in a single-sample setup, TaxVAMB on average returns 83% more high quality bins than VAMB. TaxVAMB bins incomplete genomes drastically better than any other tool, returning 255% more high quality bins of incomplete genomes than the next best binner. Our method has immediate research and industrial applications, as well as methodological novelty which can be translated to other biological problems with semi-supervised multimodal datasets.

bioinformatics↗

bioRxiv2024

Deep Learning-Based Motif Discovery in MajorHistocompatibility Complex: A Primer

In this work, we present a deep learning-based approach for motif discovery in the Major Histo-compatibility Complex (MHC) system, which plays a key role in the immune response. We explore the use of convolutional neural networks (CNNs) to identify peptide binding motifs for different MHC alleles. By training models on data from specific MHC Class I and II molecules, we demonstrate how 1-dimensional convolutional filters can effectively capture motifs and binding preferences. Our study introduces a method for extracting motif logos directly from the trained models, providing insights into how internal neural network representations align with known biological motifs. The results show significant alignment with experimental binding motifs, underscoring the utility of deep learning in immunological research and the potential for improving vaccine design and immunotherapy.

bioinformatics↗

bioRxiv2024

Sparsity of population activity in the hippocampus is task-invariant across the tri-synaptic circuit and dorsal-ventral axis

Evidence from neurophysiological and genetic studies demonstrates that activity sparsity - the proportion of neurons that are active at a given time in a population - systematically varies across the canonical trisynaptic circuit of the hippocampus. Recent work has also shown that sparsity varies across the hippocampal dorsoventral (long) axis, wherein activity is sparser in ventral than dorsal regions. While the hippocampus has a critical role in long-term memory (LTM), whether sparsity across the trisynaptic circuit and hippocampal long axis is task-dependent or invariant remains unknown. Importantly, representational sparsity has significant implications for neural computation and theoretical models of learning and memory within and beyond the hippocampus. Here we used functional molecular imaging to quantify sparsity in the rat hippocampus during performance of the Morris water task (MWT) and contextual fear discrimination (CFD) - two popular and distinct assays of LTM. We found that activity sparsity is highly reliable across memory tasks, wherein activity increases sequentially across the trisynaptic circuit (DG < CA3 < CA1) and decreases across the long axis (ventral < dorsal). These results have important implications for models of hippocampal function and suggest that activity sparsity is a preserved property in the hippocampal system across cognitive settings.

neuroscience↗

bioRxiv2024

Lower degree of microsatellite instability in colorectal carcinomas from MSH6-associated Lynch syndrome patients

BackgroundNumerous observational and molecular studies focusing on Lynch syndrome (LS) have revealed significant variation in the phenotype and molecular characteristics among carriers of pathogenic variants in mismatch repair genes (path_MMR). Recently, we demonstrated that colorectal carcinomas in path_MSH6 carriers exhibit fewer insertion/deletion mutations compared to CRCs from other MMR groups, raising the question of whether MSH6-mutated CRCs might display a lower degree of microsatellite instability (MSI). MethodsMutations at twenty coding microsatellites (cMS) were analyzed in 39 MSH6-, 18 MLH1-, 16 MSH2- and 22 PMS2-mutated CRCs and 35 sporadic MSI CRCs, and mutation frequencies and mutant allele ratios were compared among the different MMR-deficient groups. Considering factors such as HLA-A*02:01 type, B2M status, and the anticipated immunogenicity of frameshift peptides derived from cMS mutations, the identified cMS mutation profiles of MSH6-mutated CRCs were further investigated to assess their potential impact on immunotherapeutic strategies. ResultsMSH6-mutated CRCs exhibited lower mutation frequencies and mutant allele ratios across most cMS. The cMS mutations in MSH6-mutated CRCs demonstrated inverse correlations with the predicted immunogenicity of the resulting frameshift peptides, which may suggest negative selection of cell clones bearing highly immunogenic frameshift peptides. ConclusionsMSH6-mutated CRCs display a lower degree of MSI, which may be connected to lower penetrance and later onset of the disease in path_MSH6 carriers. Moreover, this lower MSI level may implicate an altered immune response compared to other MSI CRCs, which could have implications for the success of immunotherapy in MSH6-mutated CRCs. Future studies should carefully evaluate this possibility. If confirmed, these results would reinforce the notion of classifying LS as distinct syndromes associated with specific MMR genes.

genetics↗

bioRxiv2024

Bispecific GD2 x B7-H3 Antibody Improves Tumor Targeting and Reduces Toxicity while Maintaining Efficacy for Neuroblastoma

The current treatment for neuroblastoma involves an immunotherapy regimen that includes a monoclonal antibody that recognizes disialoganglioside (GD2), expressed at high levels on neuroblastoma. GD2 is not present on most normal tissues but is expressed on nerves. Thus, anti-GD2 treatment causes substantial, dose-limiting, neuropathic pain. B7-H3 is overexpressed on multiple tumor types, including neuroblastoma, with minimal normal cell expression and is absent on nerves. We designed a bispecific antibody (bsAb) that requires simultaneous binding of these two tumor antigens to achieve tight-binding of tumor cells. Our preclinical research shows that when compared to an anti-GD2 monospecific antibody, the GD2xB7-H3 bsAb has improved tumor specificity with similar efficacy and reduced toxicity. Since this bsAb does not bind to nerves, it may be possible to administer increased or additional doses beyond the tolerable dose of monospecific anti-GD2 antibodies, which could improve therapeutic efficacy and quality of life for patients with neuroblastoma.

cancer biology↗