Biology subjects

Barra, C.

Publications and source records attributed to Barra, C..

11 recordsLinked to original sources

bioRxiv2026

Chemically informed representations of amino acids enable learning beyond the canonical protein alphabet

Computational models of proteins typically represent sequences using a fixed twenty-letter alphabet describing canonical amino acids. Although this symbolic representation underlies most machine learning approaches to protein analysis, it abstracts away the chemical structure of residues and cannot naturally encode post-translational modifications (PTMs). As a result, current models struggle to incorporate chemical variation beyond the canonical amino acid alphabet. Here we introduce a chemically informed representation of peptides based on two-dimensional depictions of amino acid structures. Peptides are encoded as mosaics of residue depictions and embedded using a convolutional autoencoder, allowing machine learning models to learn physicochemical features directly from molecular structure. Because these representations capture chemical properties rather than symbolic residue identities, they enable learning across structurally related residues and support generalization to modified amino acids not explicitly observed during training. Applied to Major Histocompatibility Complex class I binding prediction, these embeddings achieve competitive performance while enabling chemically interpretable attribution of the molecular features driving predictions.

bioinformatics↗

bioRxiv2026

Resolving Genome-to-Phenotype Links in Bacteria: Machine-Learned Inference from Downsampled k-mer Representations

Standard approaches to bacterial phenotyping often treat the entire genome as the fundamental unit of information, resulting in high-dimensional inputs that may contain significant redundancy. Consequently, current bacterial phenotyping techniques typically rely on the assumption that entire sequences are required for accurate predictions. While downsampling based on min-hashing or prefix filtering has been used for clustering, its utility as a direct input for predictive machine learning remains underexplored. Here, we show that a novel prefix-based downsampling algorithm can reduce the size of genomes while maintaining relatively high predictive accuracy on phenotype prediction tasks. By combining a prefix reduction strategy with the specificity of short k-mers, we developed a method to downsample entire genomes into k-mer frequency matrices and k-mer-on-a-string representations. We found that ensemble models, such as Random Forest and Gradient Boosting, trained on k-mer frequency matrices from downsampled genome representations outperformed more complex deep learning architectures with the same downsampled representation, particularly on datasets with limited data or highly similar genomes. We were able demonstrate explainability by tracing back the k-mers with the most impact on the models to genes coding for the specific phenotype. Our results demonstrate that downsampling genomic data can yield models with good predictive power thus establishing an alternative when using full genomes is infeasible. We present an approach that offers relatively high performance on bacterial phenotyping tasks and demonstrates a path forward towards lightweight Genome Language Models that will enable analysis of entire genomes.

bioinformatics↗

bioRxiv2025

Evidence of altered antigen processing in autoimmune disease revealed by comparative immunopeptidomics

Autoimmune disease is traditionally attributed to failure of immune tolerance toward normally presented self-antigens. Although strong HLA associations implicate antigen presentation in disease risk, whether altered antigen processing contributes directly to autoimmunity remains unclear. To address this question, we analysed experimentally verified human T-cell autoantigens across twenty autoimmune diseases and classified them according to their natural presentation on HLA-DR in healthy individuals as either tolerant (detectably presented on MHC) or cryptic (not detectably presented). These classes exhibited distinct molecular and clinical associations: cryptic autoantigens were enriched in membrane proteins with tissue-specific functions and were predominantly linked to organ-specific autoimmune diseases, whereas tolerant autoantigens were largely extracellular, enriched in immune-related functions, and associated with systemic autoimmunity. Analysis of immunopeptidomics datasets from rheumatoid arthritis and multiple sclerosis further revealed disease-associated shifts in peptide flanking residues and increased relative solvent accessibility compared with healthy donors, consistent with altered proteolytic processing. Cryptic proteins were markedly overrepresented among multiple sclerosis-derived ligands. Together, these findings define two mechanistically distinct routes to autoimmune activation: one driven by altered antigen processing that exposes previously unseen self-proteins, and another driven by breakdown of tolerance toward normally presented self-antigens. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=116 SRC="FIGDIR/small/693959v2_ufig1.gif" ALT="Figure 1"> View larger version (34K): org.highwire.dtl.DTLVardef@9249f5org.highwire.dtl.DTLVardef@35a55borg.highwire.dtl.DTLVardef@1fb0beforg.highwire.dtl.DTLVardef@cc97b3_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2024

ProtParts, an automated web server for clustering and partitioning protein datasets

Data leakage originating from protein sequence similarity shared among train and test sets can result in model overfitting and overestimation of model performance and utility. However, leakage is often subtle and might be difficult to eliminate. Available clustering tools often do not provide completely independent partitions, and in addition it is difficult to assess the statistical significance of those differences. In this study, we developed a clustering and partitioning tool, ProtParts, utilizing the E-value of BLAST to compute pairwise similarities between each pair of proteins and using a graph algorithm to generate clusters of similar sequences. This exhaustive clustering ensures the most independent partitions, giving a metric of statistical significance and, thereby enhancing the model generalization. A series of comparative analyses indicated that ProtParts clusters have higher silhouette coefficient and adjusted mutual information than other algorithms using k-mers or sequence percentage identity. Re-training three distinct predictive models revealed how sub-optimal data clustering and partitioning leads to overfitting and inflated performance during cross-validation. In contrast, training on ProtParts partitions demonstrated a more robust and improved model performance on predicting independent data. Based on these results, we deployed the user-friendly web server ProtParts (https://services.healthtech.dtu.dk/services/ProtParts-1.0) for protein partitioning prior to machine learning applications. GRAPHICAL ABSTRACT O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=79 SRC="FIGDIR/small/603234v1_ufig1.gif" ALT="Figure 1"> View larger version (22K): org.highwire.dtl.DTLVardef@c0df9forg.highwire.dtl.DTLVardef@994c6borg.highwire.dtl.DTLVardef@68147eorg.highwire.dtl.DTLVardef@1198eab_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2023

Integration of HLA-DR linkagedisequilibrium to MHC class II predictions

Insights into peptide binding to HLA class II molecules is essential when studying the biological mechanisms behind cellular immunity, autoimmune diseases, and the development of immunotherapies and peptide vaccines. Currently, most of the publicly available data used to train state-of-the-art binding prediction methods for HLA-DR only includes DRB1 information. The role of the paralogue alleles, HLA-DRB3/4/5, and their strong linkage disequilibrium to DRB1 is often omitted when typing HLA-II alleles. This leads to ambiguities when making disease associations and interpreting HLA-restricted immune data. To resolve this issue, we present HLAAssoc-1.0, a method to infer HLA-DRB3/4/5 alleles by linkage disequilibrium to HLA-DRB1. We illustrate the usage of the tool and the importance of the integration of HLA-DRB3/4/5 alleles in the data analysis in different case studies including the interpretation of immunopetidomics data. Additionally, we infer allele information for the data used for training of NetMHCIIpan lacking HLA-DRB3/4/5 allele information and demonstrate that the retrained method achieved improved performance. In all cases, inferring HLA-DRB3/4/5 allele presence in non-fully typed HLA-II assays resulted in improved allele and motif deconvolutions. HLAAssoc-1.0 is available at https://services.healthtech.dtu.dk/service.php?HLAAssoc-1.0.

bioinformatics↗

bioRxiv2022

NetAllergen, a random forest model integrating MHC-II presentation propensity for improved allergenicity prediction

Allergy is a pathological immune reaction towards innocuous protein antigens. Although only a narrow fraction of plant or animal proteins induce allergy, atopic disorders affect millions of children and adults and cost billions in healthcare systems worldwide. In-silico predictors can aid in the development of more innocuous food sources. Previous allergenicity predictors used sequence similarity, common structural domains, and amino acid physicochemical features. However, these predictors strongly rely on sequence similarity to known allergens and fail to predict protein allergenicity accurately when similarity diminishes. In addition, allergen is a broad terminology that may include different compounds, hindering the classification task. To overcome these limitations, we collected allergens from AllergenOnline, a curated database of IgE-inducing allergens, carefully removed allergen redundancy with a novel protein partitioning pipeline, and developed a new allergen prediction method, introducing MHC presentation propensity as a novel feature. NetAllergen outperformed a sequence similarity-based BLAST baseline approach, and previous allergenicity predictor AlgPred 2 when similarity to known allergens is limited. NetAllergen is available as a web service (services.healthtech.dtu.dk/service.php?NetAllergen-1.0) and can predict allergenicity from a protein sequence.

bioinformatics↗

bioRxiv2022

Machine learning reveals limited contribution of trans-only encoded variants to the HLA-DQ immunopeptidome by accurate and comprehensive HLA-DQ antigen presentation prediction

HLA class II antigen presentation is key for controlling and triggering T cell immune responses. HLA-DQ molecules, which are believed to play a major role in autoimmune diseases, are heterodimers that can be formed as both cis and trans variants depending on whether the - and {beta}-chains are encoded on the same (cis) or opposite (trans) chromosomes. So far, limited progress has been made for predicting HLA-DQ antigen presentation. In addition, the contribution of trans-only variants (i.e. variants not observed in the population as cis) in shaping the HLA-DQ immunopeptidome remains largely unresolved. Here, we seek to address these issues by integrating state-of-the-art immunoinformatics data mining models with large volumes of high-quality HLA-DQ specific MS-immunopeptidomics data. The analysis demonstrated a highly improved predictive power and molecular coverage for models trained including these novel HLA-DQ data. More importantly, investigating the role of trans-only HLA-DQ variants revealed a limited to no contribution to the overall HLA-DQ immunopeptidome. In conclusion, this study has furthered our understanding of HLA-DQ specificities and has for the first time cast light on the relative role of cis versus trans-only HLA-DQ variants in the HLA class II antigen presentation space. The developed method, NetMHCIIpan-4.2, is available at https://services.healthtech.dtu.dk/services/NetMHCIIpan-4.2.

bioinformatics↗

bioRxiv2021

Accurate MHC Motif Deconvolution of Immunopeptidomics Data Reveals a Significant Contribution of DRB3, 4 and 5 to the Total DR Immunopeptidome

Mass spectrometry (MS) based immunopeptidomics is used in several biomedical applications including neo-epitope discovery in oncology and next-generation vaccine development. Immunopeptidome data are highly complex given the expression of multiple HLA alleles on the cell membrane and presence of co-immunoprecipitated contaminants. The absence of tools that accurately deal with these challenges is currently a major bottleneck for the large-scale application of this technique. Here, we present the MHCMotifDecon that benefits from state-of-the-art HLA class-I and class-II predictions to accurately deconvolute immunopeptidome datasets and assign individual ligands to the most likely HLA allele while discarding co-purified contaminants. We have benchmarked the tool against other state-of-the-art methods and illustrated its application on experimental datasets for HLA-DR demonstrating a previously underappreciated role for HLA-DRB3/4/5 molecules in defining HLA class II immune repertoires. With its ease of use MHCMotifDecon can efficiently guide interpretation of immunopeptidome datasets, serving the discovery of novel T cell targets.

immunology↗

bioRxiv2019

Improved prediction of MHC II antigen presentation throughintegration and motif deconvolution of mass spectrometryMHC eluted ligand data

Major Histocompatibility Complex II (MHC II) molecules play a vital role in the onset and control of cellular immunity. In a highly selective process, MHC II presents peptides derived from exogenous antigens on the surface of antigen-presenting cells for T cell scrutiny. Understanding the rules defining this presentation holds critical insights into the regulation and potential manipulation of the cellular immune system. Here, we apply the NNAlign_MA machine learning framework to analyse and integrate large-scale eluted MHC II ligand mass spectrometry (MS) data sets to advance prediction of CD4+ epitopes. NNAlign_MA allows integration of mixed data types, handling ligands with multiple potential allele annotations, encoding of ligand context, leveraging information between data sets, and has pan-specific power allowing accurate predictions outside the set of molecules included in the training data. Applying this framework, we identified accurate binding motifs of more than 50 MHC class II molecules described by MS data, particularly expanding coverage for DP and DQ beyond that obtained using current MS motif deconvolution techniques. Further, in large-scale benchmarking, the final model termed NetMHCIIpan-4.0, demonstrated improved performance beyond current state-of-the-art predictors for ligand and CD4+ T cell epitope prediction. These results suggest NNAlign_MA and NetMHCIIpan-4.0 are powerful tools for analysis of immunopeptidome MS data, prediction of T cell epitopes and development of personalized immunotherapies.

bioinformatics↗

bioRxiv2019

NNAlign_MA; semi-supervised MHC peptidome deconvolution for accurate characterization of MHC binding motifs and improved T cell epitope prediction

Antigen presentation by Major Histocompatibility Complex (MHC) is a cornerstone of the cellular immune system, and characterization of the rules defining which peptides are suitable for MHC presentation is of key interest for understanding T cell immunogenicity and the development of T cell driven therapeutics. The collective set of peptides presented on a cell surface by MHC molecules is known as the immunopeptidome. Due to the vast MHC polymorphism, this peptidome is unique between individuals. Current state-of-the-art liquid chromatography mass spectrometry (LC-MS) technologies allow the identification of large peptidomes specific for a given host or cell line, and numerous studies have proven this data to be a rich source of information for learning the rules of MHC-mediated antigen presentation. The data obtained with MS techniques is usually poly-specific - i.e. it contains multiple sequence motifs matching the different MHC molecules expressed in the system under investigation. Deconvolution of this poly-specificity has proven a challenge, and imposes a substantial limitation for learning MHC specific rules for antigen presentation from immunopeptidome data. This since each ligand first must be unambiguously associated to its presenting MHC molecule(s) within the haplotype of the cell line. Here, we describe NNAlign_MA, a method that is able to address this challenge. NNAlign_MA is capable of taking a mixed training set composed of single-allele (SA) data (peptides assigned to single MHCs) and multi-allele (MA) data (peptides with multiple options for MHCs assignments) as input, and in a fully automated manner deconvolute the individual MHC restriction of all MA sequences while learning the binding specificities of all the MHCs present in the training set. The NNAlign_MA method was benchmarked on a large and diverse dataset, covering human MHC class I and class II, and bovine class I (BoLA) data. For all cases, the method was demonstrated to have unprecedented performance compared to state-of-the-art methods, achieving a complete deconvolution of binding motifs contained within poly-specificity MS eluted ligand data and an improved predictive performance for identification of both eluted ligands and T cell epitopes. Given its very high flexibility and ease of use, we expect NNAlign_MA to serve as an effective tool to increase our understanding of the rules for MHC antigen presentation and guide the development of novel T cell-based therapeutics.

bioinformatics↗

bioRxiv2017

Computational Tools for the Identification and Interpretation of Sequence Motifs in Immunopeptidomes

Recent advances in proteomics and mass-spectrometry have widely expanded the detectable peptide repertoire presented by major histocompatibility complex (MHC) molecules on the cell surface, collectively known as the immunopeptidome. Finely characterizing the immunopeptidome brings about important basic insights into the mechanisms of antigen presentation, but can also reveal promising targets for vaccine development and cancer immunotherapy. In this report, we describe a number of practical and efficient approaches to analyze immunopeptidomics data, discussing the identification of meaningful sequence motifs in various scenarios and considering current limitations. We address the issue of filtering false hits and contaminants, and the problem of motif deconvolution in cell lines expressing multiple MHC alleles, both for the MHC class I and class II systems. Finally, we demonstrate how machine learning can be readily employed by non-expert users to generate accurate prediction models directly from mass-spectrometry eluted ligand data sets.

bioinformatics↗