Siarheyeu, R. | Search bioRxiv

bioRxiv2026

Machine Learning-Driven Antigen Selection Reveals Conserved T-Cell Targets for Broad Coronavirus Vaccination

BackgroundCoronavirus outbreaks remain a persistent threat to global health, and vaccines based primarily on spike-specific immune responses are susceptible to antigenic variation. T-cell immunity directed against conserved internal viral proteins may provide a complementary and more variant-tolerant strategy for next-generation coronavirus vaccines. MethodsWe combined machine learning-guided antigen prioritization with ex vivo functional immunological validation to identify conserved non-spike T-cell targets across betacoronaviruses. Candidate sequences were screened for immunogenicity using primary human peripheral blood mononuclear cells from healthy donors using intracellular cytokine staining and activation-induced marker assays. Top-ranked conserved regions were incorporated into multiepitope mRNA constructs, and their intracellular expression and HLA class I presentation were confirmed by immunopeptidomics. Immunogenicity was further evaluated ex vivo and in vivo using mRNA immunization of mice and T-cell FluoroSpot assays. FindingsAcross a panel of 97 peptides derived from 19 viral proteins, evolutionary conservation across distinct betacoronavirus taxa was strongly associated with functional T-cell immunogenicity in human donors. Highly conserved peptides elicited significantly stronger and more frequent CD4 and CD8 T-cell responses than taxon-restricted peptides. Multiepitope mRNA constructs encoding conserved regions were efficiently expressed and presented on HLA class I molecules and induced T-cell responses in human PBMCs. In mice, mRNA immunization with conserved multiepitope constructs generated robust interferon-{gamma}- and interleukin-2-producing T-cell responses that exceeded those induced by unconserved control constructs. InterpretationThese results link evolutionary conservation to functional cellular immunogenicity and demonstrate the feasibility of multiepitope mRNA delivery for inducing conserved coronavirus-directed T-cell responses. Although protective efficacy remains to be established, conservation-guided antigen selection represents a scalable strategy for developing T-cell-focused vaccines with broad lineage coverage, supporting pandemic preparedness beyond spike-centered immunity. FundingThe research was supported by CEPI, NEC, University of Oslo and Oslo university hospital. Research in contextO_ST_ABSEvidence before this studyC_ST_ABSPrior coronavirus vaccine development has focused predominantly on spike protein-directed neutralizing antibodies. While highly effective against matched strains, spike-centered immunity is vulnerable to antigenic drift and lineage-specific escape. Multiple observational and experimental studies have shown that T-cell responses, particularly against internal viral proteins, are more conserved and correlate with reduced disease severity and cross-variant recognition. Epitope prediction algorithms and immunoinformatics approaches have been widely used to nominate candidate T-cell targets; however, systematic functional validation of conserved non-spike antigens across betacoronaviruses in primary human immune systems, combined with antigen presentation data and in vivo vaccine testing, has remained limited. Searches of PubMed and bioRxiv up to December 2025 using terms including "coronavirus T-cell vaccine," "conserved coronavirus epitopes," "betacoronavirus cross-reactive T cells," and "mRNA T-cell vaccine" identified studies demonstrating cross-reactive T-cell immunity and computational epitope selection, but few integrated machine-learning-guided antigen prioritization with ex vivo human functional screening, immunopeptidomics, and in vivo mRNA immunization in a unified workflow. Added value of this studyThis study provides an integrated experimental and computational framework for identifying and validating conserved non-spike T-cell antigens across betacoronaviruses. We functionally screened a panel of candidate peptides derived from multiple viral proteins and demonstrated that evolutionary conservation across species is strongly associated with T-cell immunogenicity. We further demonstrate that multiepitope mRNA constructs encoding these top-ranked conserved regions can be intracellularly expressed, presented on HLA class I molecules to induce polyfunctional T-cell responses in primary human PBMCs. Finally, in vivo mRNA immunization in mice induces robust interferon-{gamma} and interleukin-2 T-cell responses exceeding those induced by unconserved control constructs. Together, these findings link evolutionary conservation to functional cellular immunogenicity and extend beyond in silico prediction by demonstrating antigen processing, presentation, and immunogenicity across human and murine systems. Implications of all the available evidenceCollectively, the available evidence indicates that T-cell immunity directed toward conserved internal coronavirus proteins represents a complementary and potentially more variant-tolerant axis of vaccine design than spike-only strategies. Our findings suggest that evolutionary conservation can serve as a practical selection principle for prioritizing T-cell antigens with broad lineage coverage and that multiepitope mRNA delivery is a feasible platform for inducing such responses. While direct protection and heterologous challenge studies will be required to establish clinical efficacy, the integration of computational prioritization with functional validation supports a scalable approach to pandemic preparedness that may be applicable to other rapidly evolving viral families.

immunology↗

bioRxiv2021

Microbiome-based disease prediction with multimodal variational information bottlenecks

Scientific research is shedding light on the interaction of the gut microbiome with the human host and on its role in human health state. Existing machine learning methods have shown great potential in discriminating healthy from diseased microbiome states. Most of them leverage shotgun metagenomic sequencing to extract gut microbial species-relative abundances or strain-level markers. Each of these gut microbial features showed diagnostic potential when tested separately; however, no existing approach combines them in a single predictive framework. Here, we propose the Multimodal Variational Information Bottleneck (MVIB), a novel deep learning model capable of learning a joint representation of multiple heterogeneous data modalities. MVIB achieves competitive classification performance while being faster than existing methods. Additionally, MVIB offers interpretable results. Our model adopts an information theoretic interpretation of deep neural networks and computes a joint stochastic encoding of different input data modalities. We use MVIB to predict whether human hosts are affected by a certain disease by jointly analysing gut microbial species-relative abundances and strain-level markers. MVIB is evaluated on human gut metagenomic samples from 11 publicly available disease cohorts covering 6 different diseases. We achieve high performance (0.80 < ROC AUC < 0.95) on 5 cohorts and at least medium performance on the remaining ones. We adopt a saliency technique to interpret the output of MVIB and identify the most relevant microbial species and strain-level markers to the models predictions. We also perform cross-study generalisation experiments, where we train and test MVIB on different cohorts of the same disease, and overall we achieve comparable results to the baseline approach. Further, we evaluate our model by adding metabolomic data derived from mass spectrometry as a third input modality. Our method is scalable with respect to input data modalities and has an average training time of < 1.4 seconds. The source code and the datasets used in this work are publicly available. Author summaryThe gut microbiome can be an indicator of various diseases due to its interaction with the human system. Our main objective is to improve on the current state of the art in microbiome classification for diagnostic purposes. A rich body of literature evidences the clinical value of microbiome predictive models. Here, we propose the Multimodal Variational Information Bottleneck (MVIB), a novel deep learning model for microbiome-based disease prediction. MVIB learns a joint stochastic encoding of different input data modalities to predict the output class. We use MVIB to predict whether human hosts are affected by a certain disease by jointly analysing gut microbial species-relative abundance and strain-level marker profiles. Both of these gut microbial features showed diagnostic potential when tested separately in previous studies; however, no research has combined them in a single predictive tool. We evaluate MVIB on various human gut metagenomic samples from 11 publicly available disease cohorts. MVIB achieves competitive performance compared to state-of-the-art methods. Additionally, we evaluate our model by adding metabolomic data as a third input modality and we show that MVIB is scalable with respect to input feature modalities. Further, we adopt a saliency technique to interpret the output of MVIB and identify the most relevant microbial species and strain-level markers to our model predictions.

bioinformatics↗