scMOVIR: a single-cell multi-omics database for human viral infections and immune responses

Abstract

Viral infections impose a substantial threat to human health, characterized by a wide range of pathogens, clinical manifestations, and complex immune responses. Single-cell multi-omics technologies have revolutionized the study of antiviral immunity by resolving cellular heterogeneity, transcriptional reprogramming, and clonal dynamics. However, no resource has yet comprehensively integrated such datasets in the context of viral infections. Here, we present scMOVIR, a single-cell multi-omics database for human viral infections and immune responses. The database systematically integrates transcriptomic, proteomic, and immune receptor repertoire profiles at single-cell resolution, compiling 8532 human samples across 114 viral species, subtypes, and strains. These datasets span 59 well-defined disease states, including acute and chronic infections, virus-associated malignancies, and immune-related disorders, and extend across >30 tissue types. In addition, these datasets incorporate vaccination cohorts, therapeutic interventions, and time-course models of infection. To ensure consistency and interoperability, all datasets undergo standardized preprocessing, including batch correction and unified cell-type annotation, with metadata harmonized using authoritative ontologies. scMOVIR provides user-friendly interfaces for dataset exploration and interactive visualization of cellular dynamics and molecular profiles, offering a high-resolution reference for investigating virus–host immune interactions and supporting antiviral research. The database is freely accessible at https://pgx.zju.edu.cn/scmovir.

Graphical Abstract

Graphical Abstract.

Introduction

Viral infections represent a major threat to human health, encompassing diverse pathogens and heterogeneous clinical manifestations ranging from acute self-limiting infections to persistent or latent infections [1–3]. Epidemiological studies estimate that over 12% of newly diagnosed cancers are attributable to oncogenic viral infections, with an even higher proportion in resource-limited regions [4–6]. For example, human immunodeficiency virus (HIV) establishes a persistent infection characterized by chronic immune activation, progressive CD4⁺ T cell depletion, and compromised host defense [7]. Hepatitis B virus (HBV) and human papillomavirus are strongly associated with malignancies such as hepatocellular carcinoma and cervical cancer [8–10]. Epstein–Barr virus (EBV) is linked to multiple lymphoid malignancies and exhibits complex latency and reactivation cycles [11]. Additionally, the outbreak of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has led to a global pandemic, resulting in millions of deaths worldwide [12, 13]. Therefore, elucidating virus–host immune interactions is critical to advancing our understanding of viral pathogenesis and its clinical implications.

Viral infections elicit dynamic and spatially compartmentalized immune responses characterized by activation of distinct cellular subsets, phenotypic remodeling, and antigen-driven clonal expansion [14–16]. These responses are accompanied by transcriptional reprogramming and signaling network rewiring, collectively reshaping cellular functional states [16, 17]. Conventional bulk omics approaches, which average signals across heterogeneous cell populations, are insufficient to resolve these complexities [18–20]. Single-cell multi-omics technologies provide the necessary resolution to capture cellular heterogeneity and dynamic molecular responses during viral infection [21, 22]. For example, paired single-cell RNA sequencing (scRNA-seq) and T-cell receptor sequencing (scTCR-seq) in SARS-CoV-2 infection have revealed rapid antigen-specific T-cell expansion and interferon-driven mucosal immune responses during early infection [23]. In HBV-associated hepatocellular carcinoma, multi-omics profiling identified exhausted and cytotoxic CD8⁺ T cell subsets associated with distinct tumor recurrence patterns [24]. Similarly, studies of HIV and EBV infections applied single-cell multi-omics to delineate virus-specific clonal hierarchies and immune dysregulation pathways [25, 26]. These advances underscore the critical role of single-cell technologies in elucidating the cellular and molecular basis of antiviral immunity and pathogenesis.

Several publicly available databases have been developed. For example, MVIP offers extensive bulk multi-omics datasets alongside a subset of scRNA-seq data on virus–host interactions [27], and VThunter provides single-cell resolution profiles of viral receptor expression across human tissues to assess tissue susceptibility [28]. Other platforms, such as TISCH2 [29] and SC2Disease [30], emphasize the immune landscape in cancer and other diseases based on single-cell transcriptomic profiles. More recently, scImmomics has integrated T-cell receptor and B-cell receptor (TCR/BCR) repertoires with scRNA-seq data to reconstruct immune cell lineage relationships [31], and additional resources further expand specific aspects of viral single-cell immunology [32, 33]. While these resources have greatly contributed to the field, the inherent complexity of antiviral immune responses, such as rapid clonal dynamics, systemic immune activation, and pathogen-specific signatures, remains challenging to fully elucidate through bulk data. Moreover, existing single-cell resources primarily focus on transcriptomic modalities within oncological contexts, with limited coverage of diverse viral species and infection stages. In addition, inconsistencies in data preprocessing and annotation standards across studies hinder seamless integration and comprehensive cross-study analysis. These limitations underscore the need for a unified single-cell multi-omics platform specifically designed for human viral infections and immune responses.

Therefore, we developed scMOVIR, a multi-omics database for human viral infections and immune responses, which integrates transcriptomic, proteomic, and immune receptor repertoire data at single-cell resolution. scMOVIR incorporates 8532 samples covering 114 viral species, subtypes, and strains, 59 disease states, and over 30 tissue types, spanning acute and chronic infection contexts, virus-associated cancers, vaccination cohorts, and therapeutic intervention studies. All datasets are subjected to standardized processing, quality control (QC), and unified annotation to ensure consistency and interoperability. We performed multi-layered analyses, including differential gene and protein expression profiling, functional enrichment, cell–cell communication inference, and clonality characterization. These results are systematically integrated into the database and accessible through flexible, interactive visualization tools. Compared with existing resources, scMOVIR provides a comprehensive, virus-focused single-cell multi-omics platform with standardized datasets and comparative analysis, enabling in-depth investigation of virus–host interactions and immune response heterogeneity across diverse infection contexts (Supplementary Table S1).

Materials and methods

Systematic collection and rigorous curation of datasets

To establish a comprehensive and high-quality collection of single-cell datasets related to viral infections, we implemented a multi-step data acquisition and management strategy: (i) Comprehensive literature review: We conducted exhaustive searches of public virus repositories, including ViPR [34], ICTV [35], and NCBI Virus [36], to identify viruses known to infect humans. Subsequently, a systematic literature search was conducted on PubMed using Entrez E-utility, employing keyword combinations such as “single-cell RNA sequencing + virus name,” “single-cell proteomics + virus name,” and “single-cell BCR/TCR + virus name.” The search, completed in December 2024, yielded 3316 relevant publications. (ii) Public database mining: To complement the literature-based search, we conducted targeted mining of public omics repositories, including the Gene Expression Omnibus (GEO) [37], ArrayExpress [38], FlowRepository [39], SingPro [40], and SPDB [41], among others. This effort identified an additional 58 experiments covering various omics modalities. (iii) Verification and validation: All retrieved publications and datasets undergo rigorous validation through manual inspection of original articles and supplementary materials. Only publicly available single-cell datasets with clearly confirmed viral infection status were included. Comprehensive metadata were manually curated to ensure consistency and completeness. (iv) Data deduplication and integration: To eliminate redundancy, duplicated datasets were manually reviewed and consolidated. In cases where samples overlap across multiple studies, the dataset with more complete metadata is retained, and annotations are added to indicate the source of the original data to ensure data traceability.

Data quality control and preprocessing

scMOVIR employs a modality-specific standardized preprocessing pipeline to ensure data quality and enable cross-study comparability [42]. For single-cell transcriptomic data, we implemented a two-step QC strategy that combines empirical thresholds with sample-level median absolute deviation (MAD)-based outlier detection for all datasets. QC was performed using Scanpy [43]. First, low-quality cells were explicitly checked and removed based on established empirical thresholds: (i) cells with low gene counts (n_genes <200), (ii) cells with low total counts (total_counts <500), and (iii) cells with high mitochondrial content (pct_counts_mt >10%). Then, sample-level outlier detection was performed based on the MAD of four metrics [44]: log-transformed gene counts, log-transformed total transcript counts, proportion of transcripts from the top 20 most highly expressed genes, and mitochondrial transcript fraction. MAD thresholds of 5 were applied to the first three metrics and 3 to the mitochondrial proportion.

For droplet-based single-cell sequencing, Scrublet [45] was applied to each sample to detect and remove potential doublets. The algorithm assigns a doublet score to each cell by simulating synthetic doublets and comparing them to observed profiles. Cells exceeding the score threshold were excluded using the following parameters: min_counts = 2, min_cells = 3, and min_gene_variability_pctl = 85. For each dataset, genes detected in fewer than three cells were filtered out. Expression counts were then normalized by total counts per cell (TCM), scaled to 10,000 counts, and log-transformed using log1p. Highly variable genes (HVGs) were identified using a dispersion-based selection method with the following parameters: min_disp = 0.5, min_mean = 0.0125, max_mean = 3, span = 0.3, and n_bins = 20. Principal component analysis (PCA) was subsequently performed on the selected HVGs using the ARPACK solver to compute 50 principal components. Batch effects were corrected using Harmony [46], with sample ID specified as the batch key. Nearest-neighbor graphs were constructed from the Harmony-corrected PCA embeddings. UMAP was applied for dimensionality reduction and visualization, using Euclidean distance with n_neighbors = 15, min_dist = 0.5, and spread = 1.0. Cell clustering was performed using the Leiden algorithm with an initial resolution of 1, which was further adjusted during manual cell-type annotation to resolve subpopulation structures more accurately.

For single-cell proteomic data from cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) experiments, cells with low counts, determined based on the distribution of captured antibody-derived tags, were removed. Sample-level QC was performed using MAD-based filtering on detected gene counts and total UMI counts with a 5-MAD cutoff. Normalization was conducted via centered log-ratio transformation. For flow cytometry-based proteomic datasets, raw data were imported into FlowJo (v 10.8) for QC, including removal of anomalous events based on scatter and signal intensity. Dead cells and atypical populations were manually gated out, and signal intensities were normalized using arcsinh transformation.

For scTCR/BCR-seq data, processing was performed using Scirpy [47]. Only cells with paired productive receptor sequences were retained: T cells with both α-chain (TRA) and β-chain (TRB), and B cells with both heavy (IGH) and light (IGK or IGL) chains. Clonotypes were defined as unique paired TRA–TRB or IGH–IGK/IGL sequence combinations, with clonal events required to be detected in at least two cells. For each clonotype, clonal frequency per cell, the proportion of clonotype-positive cells within each cell type, and clonotype distribution across cell types were calculated.

Cell-type annotation

We have implemented a combined automated and manual curation strategy. Initially, CellTypist was used to automatically annotate each dataset [48], employing the pretrained Immune_All_Low.pkl model. Predictions were first made at the single-cell level, followed by a majority-vote-based consensus to assign robust cell type labels and mitigate local over-clustering artifacts. For clusters with low prediction confidence scores, we performed manual inspection using a curated list of marker genes validated in the literature, and manually corrected annotations for clusters that were inconsistent with expected marker gene expression. Supplementary Table S2 provides a comprehensive list of cell types, subtypes, and the marker genes collected from literatures [49–52]. For datasets that were derived from cell lines or specific cell types, manual curation was conducted based on literature and associated metadata.

Differentially expressed genes and proteins

Differentially expressed genes (DEGs) and proteins were identified using Welch’s t-test on normalized, log-transformed expression data. DEGs were selected using false discovery rate <0.05, absolute log₂ fold change >1, and gene expression in >20% of cells within the cluster. Cluster signature genes were defined as those significantly upregulated relative to all other clusters and expressed in the majority of cells within the cluster.

Functional enrichment analysis

Functional enrichment analysis was conducted using over-representation analysis. DEGs with adjusted P-values <.05 were assessed for enrichment of Gene Ontology (GO) biological processes [53, 54], Reactome [55], and KEGG pathways using Fisher’s exact test with Benjamini–Hochberg correction [56, 57]. Only pathways meeting the corrected P-value threshold (<.05) were considered significant. The entire workflow was implemented via GSEApy [58].

Cell–cell communication

Cell–cell communication networks were inferred using CellphoneDB [59], which predicts ligand–receptor interactions between cell types based on scRNA-seq expression data. For each condition group, 1000 empirical permutations were performed by randomly shuffling cell cluster labels to estimate the null distribution of average ligand–receptor expression within interacting clusters. Ligand–receptor pairs with adjusted P-values <.05 were considered significant. To identify differential interactions, the log₂ fold change (log₂FC) of ligand–receptor communication scores between the disease/virus group and the control group was calculated for each interacting cell pair, enabling the systematic detection of intercellular interactions that are specifically altered under experimental or infection conditions.

Clonal diversity scoring

Clonal diversity was assessed by computing the Shannon entropy of clonotype frequencies within each sample. To account for variability in sequencing depth, the observed entropy was normalized by dividing by its theoretical maximum, yielding a normalized diversity index ranging from 0, where a single clonotype dominates, to 1, representing maximal clonotype diversity. This approach enables quantitative comparison of clonal heterogeneity across samples.

Database implementation

scMOVIR is built on a modular architecture, with the backend developed using Django (v3.2.16) and MySQL (v5.7.34) for data management. The frontend utilizes Vue.js (v3.5.1) combined with Element Plus components to deliver an interactive and user-friendly interface. Visualization functionalities are implemented using ECharts (v5.5.1) [60] and Highcharts (v11.4.8), enabling dynamic rendering of gene expression profiles, high-dimensional single-cell data distributions, and immune receptor repertoires, thereby supporting comprehensive multi-omics data exploration.

Factual content and data retrieval

Overview of scMOVIR

The overall framework of scMOVIR is illustrated in Fig. 1. Through systematic collection from GEO [37], PubMed, and other related repositories, scMOVIR is a rigorously curated, systematically organized database that integrates high-quality single-cell multi-omics datasets focused on human viral infections and immune responses. The database comprises 8532 samples from 352 independent experiments, encompassing single-cell transcriptomics, single-cell proteomics (including scCITE-seq and fluorescence-based flow cytometry), and single-cell immune repertoire data, which includes paired TCR and BCR sequencing data (Fig. 1A). These datasets comprehensively capture the cellular and molecular landscape of virus–host interactions, encompassing 114 viral species, subtypes, and strains, including major pathogens such as HIV, HBV, and SARS-CoV-2. This collection covers 59 disease states, spanning acute and chronic infections, virus-associated malignancies (such as cervical cancer and hepatocellular carcinoma), and virus-associated immune dysregulation (such as lymphocytopenia and systemic lupus erythematosus), across >30 tissue types, including lung, liver, blood, lymph nodes, and brain. In addition, these datasets incorporate vaccination cohorts, clinical treatment interventions, longitudinal infection dynamics, and in vitro viral perturbation experiments.

Figure 1.

Overview of the scMOVIR workflow. (A) Systematic collection of single-cell multi-omics datasets related to viral infections from literature and multiple public repositories. All datasets undergo stringent QC and preprocessing, supplemented by manual curation and comprehensive metadata harmonization to ensure consistency and accuracy. (B) Downstream analyses and visualization are performed for each omics modality. Transcriptomic analyses encompass cell UMAP projection, cell-type composition, differential gene expression, functional enrichment, and cell–cell communication. Proteomic analyses involve the identification of differentially expressed proteins. Immune repertoire analyses include quantification of clonal abundance, repertoire diversity, and VJ gene usage patterns.

All datasets undergo stringent QC, standardized preprocessing, and harmonized downstream processing, ensuring analytical robustness and cross-cohort comparability. The scMOVIR platform provides a unified framework for systematic interrogation of virus–host interactions, delineation of immune response dynamics, and mechanistic insights into viral pathogenesis across diverse clinical and experimental contexts.

Intuitive and structured user interface

The scMOVIR platform offers a user-friendly and structured interface to facilitate efficient exploration of single-cell datasets related to viral infections and associated diseases. Users can access the data through different modules, including Browser, Search, Analysis, and Download (Fig. 2). Within the “Browse” module, users can access an interactive portal for data navigation (Fig. 2A). The left panel categorizes viruses and diseases, incorporating a hierarchical classification system. Viruses are organized according to the Baltimore classification, while diseases follow ICD-11 standards, including acute infection, chronic disease, cancer, immune dysfunction, vaccination, sequelae, and others. Selecting a subcategory displays card summaries on the right of relevant entries, each summarizing the scientific name, identifier, encompassed omics types, and experiment counts. Users can click the “Detail” button to navigate to the detail page, which presents the relationship between the selected virus and associated diseases, tissue involvements, and all relevant datasets.

Figure 2.

Main modules and functions of scMOVIR. (A) The Browse page provides an interactive portal to explore viruses and diseases, organized by the Baltimore classification and the ICD-11 system, respectively. Users can select categories to access summary cards with key information and navigate to detailed pages that present virus–disease relationships and associated datasets. (B) The Search page supports keyword queries across multiple omics datasets, including virus names, diseases, tissues, viral strains, experimental models, and specific proteins for proteomics data. Search results are displayed in both list and table formats, summarizing key information of single-cell experiments. (C) The experiment details page provides comprehensive metadata, including experimental design, sample and batch information, related experiments and datasets, analytical outcomes, interactive visualizations, downloadable files, and others. (D) The analysis and visualization module comprises a suite of interactive, high-resolution visualizations tailored to each omics layer. All plots are dynamically linked to metadata and results, enabling multi-scale interrogation of viral infection-associated cell states, molecular profiles, and intercellular communication. (E) The database incorporates two analytical tools: the Gene Signature Analysis module, which evaluates enrichment of user-defined gene sets in viral infection-related conditions, and the Integrated Viral Infection Atlas module, which enables integrative analysis of single-cell transcriptomic landscapes across various viral infections.

scMOVIR offers a search module with autocompletion, enabling queries across various omics datasets based on “virus name,” “disease name,” and “tissue name” (Fig. 2B). All search terms are standardized and linked to authoritative databases to ensure consistency. In the context of single-cell transcriptomics, datasets derived from in vitro viral infections can also be retrieved using viral strain names and experimental models (e.g. cell lines, organoids). For proteomics data, queries can be performed based on specific protein markers. Search results are presented in both list and table formats, summarizing key information for each single-cell experiment. Users can access detailed views of individual experiments, supporting in-depth analysis and data download.

Comprehensive experimental metadata and data accessibility

Comprehensive and standardized metadata are critical for ensuring transparency, reproducibility, and cross-study integration in large-scale single-cell research. Therefore, scMOVIR systematically curates all experiments, providing detailed information, including experimental metadata, sample list of characteristics, similar experiments, and downloadable files (Fig. 2C). The upper section of each experiment page presents metadata including project identifiers, study titles, and reference links to original publications (Fig. 3A). Project descriptions and experimental design details, such as study objectives, grouping strategies, library preparation (e.g. 10× Chromium 3′, 10× Chromium 5′, SMART-seq2), sequencing platforms (e.g. Illumina NovaSeq 6000), number of samples, and processed single-cells, are fully documented. Standardized biological context information is provided, such as viral species/subtypes, infection status, disease classification, and tissue origin. Furthermore, the complete data processing pipeline is also recorded, specifying QC procedures, normalization methods, batch correction strategies, clustering algorithms, and all software versions and parameter settings. Each single-cell experiment is provided with a detailed sample characteristics list containing sample identifier, title, source tissue, patient information, and experimental variables (e.g. infection status), along with technical batch annotations to support robust stratification and reproducibility (Fig. 3B). For proteomic datasets, antibody panel specifications, including target antigens, labeling strategies (fluorescent or isotopic), and detection modalities, are fully documented (Fig. 4A). Immune receptor datasets are annotated with paired TCR/BCR chain sequences, clonotype assignments, CDR3 characteristics, and VJ gene usage profiles, enabling precise delineation of antigen-specific immune repertoires (Fig. 4B).

Figure 3.

Representative scMOVIR detail page illustrating general information for a single-cell experimental dataset. (A) The upper section of each project page displays metadata, including project identifier, project title, virus type, disease classification, tissue origin, study description, experimental design details, data processing, and reference to the original publication. (B) The sample list displays sample identifiers, sample titles, source tissues, patient information, experimental variables, and batch information, supporting reliable stratified analysis and ensuring experimental reproducibility.

Figure 4.

Representative scMOVIR page describing the protein panel and immune receptor information. (A) Proteomics datasets provide detailed antibody panel specifications, encompassing target antigens, labeling strategies (fluorescent or isotopic), and detection modalities. (B) Immune receptor datasets are annotated with paired TCR/BCR chain sequences, clonotype assignments, VJ gene usage, and CDR3 sequences, enabling precise characterization of antigen-specific immune repertoires.

Furthermore, scMOVIR establishes systematic links between related experiments associated with the same viral infections, enabling comprehensive navigation and interrogation of infection-specific data within a unified metadata framework. For each experiment, the database provides downloadable files, including standardized annotations for cell types, proteomic profiles, and immune receptor repertoires, supporting cell classification, immune profiling, and multi-omics integration. Analytical results, including differential expression, cell-type abundance, and functional enrichment, are also made available to facilitate further data mining and validation.

Interactive visualization of cellular and molecular features

scMOVIR provides a variety of interactive visualizations to facilitate in-depth exploration of cellular heterogeneity, molecular signatures, intercellular communication, and immune repertoire diversity in the context of viral infection (Figs 1B and 2D). For single-cell transcriptomics, annotated cell populations are visualized using UMAP projections based on clustered gene expression profiles, which intuitively represent the complex spatial relationships between clusters and cell types. For datasets exceeding 5000 cells, a downsampled subset is displayed to ensure efficient web rendering while preserving the overall data structure and cell type proportions. Pie charts summarize the relative abundance of different cell populations within each reference dataset, offering an overview of cellular composition. Boxplots combined with statistical tables enable quantitative comparisons of cell population distributions across experimental groups, facilitating the identification of infection- or phenotype-specific cellular remodeling. The heatmaps display the most significantly differentially expressed genes across cell types, providing an intuitive visualization of expression patterns and cell type-specific features. Differential gene expression analysis identifies genes significantly up- or downregulated in defined cell types and experimental conditions, with functional interpretation supported by enrichment analysis using GO, KEGG, and Reactome. Enrichment results are presented as bubble plots integrating gene counts and statistical significance. Intercellular communication is inferred through ligand–receptor interaction networks, visualized as circular plots of intracellular and intercellular signaling pathways, allowing users to interrogate major signal sources and targets, communication probabilities, and top-ranked ligand–receptor pairs that reflect infection-induced immune regulatory circuits. Differential signaling results are further highlighted with a scatter plot, which depicts the most significant interactions between cell types and illustrates both the magnitude and specificity of cell–cell communication changes across disease groups.

For single-cell proteomics, differential protein analysis is visualized using boxplots and numerical summaries for specific markers, capturing infection-associated changes in protein abundance. For immune receptor analysis, the normalized Shannon entropy index quantifies clonal diversity, repertoire evenness, and oligoclonal expansion. Histograms of CDR3 length distributions reveal clonal expansion patterns and selection pressures, reflecting structural biases in antigen recognition. Heatmaps depict V(D)J gene rearrangements and usage preferences, highlighting remodeling of antigen-specific immune receptors at the gene segment level. Collectively, these interactive visualizations enable comprehensive characterization of cellular states, molecular remodeling, intercellular signaling, and adaptive immune receptor dynamics during viral infection.

Integrated analytical functions in scMOVIR

On the analysis page, scMOVIR offers two analytical modules: Viral Infection Gene Signature Analysis and the Integrated Viral Infection Atlas, designed to facilitate systematic investigation of host–virus interactions at single-cell resolution (Fig. 2E). The first tool assesses whether user-defined genes are associated with viral infection-related diseases or phenotypes and systematically evaluates their enrichment patterns. For example (Fig. 5A), upon submission of a gene list comprising inflammatory cytokines and chemokines (e.g. IL6, IL1B, TNF, CXCL10, CXCL9, and CCL2), the system calculates enrichment significance across viral infection-related datasets using Fisher’s exact test. The outputs include a gene–virus association map illustrating differential expression of the submitted genes across viral infection contexts, an enrichment overview across disease states highlighting infection-specific patterns, and differential expression statistics presented as boxplots with accompanying tables. Collectively, these results enable identification of infection-associated gene signatures and provide a robust reference for cross-dataset comparisons and functional validation.

Figure 5.

Analysis tools available in scMOVIR. (A) Viral Infection Gene Signature Analysis assesses associations between user-defined gene sets and viral infection-related diseases or phenotypes. Given an input gene list (e.g. inflammatory cytokines IL6, IL1B, TNF, CXCL10, CXCL9, CCL2), enrichment analysis across viral datasets is performed using Fisher’s exact test. The output comprises a gene–virus association map, an enrichment overview stratified by disease state, and differential expression statistics, thereby enabling the identification of infection-associated signatures and facilitating cross-dataset comparisons. (B) Integrated Viral Infection Atlas aggregates large-scale, patient-derived single-cell RNA sequencing datasets to delineate immune cell remodeling and transcriptional dynamics during viral infection. Using HIV as an example, disease annotations from all related experiments are harmonized, and the scVI model is applied to construct a reference framework anchored to healthy controls. The output provides cell type annotations, cell composition profiles, differential gene expression, and pathway enrichment analyses, offering mechanistic insights into specific virus–host interactions.

The second module integrates large-scale, patient-derived viral infection scRNA-seq datasets to characterize immune cell state remodeling and transcriptional reprogramming during infection and disease progression. As an example, for HIV infection (Fig. 5B), scMOVIR aggregates all HIV-related scRNA-seq datasets, harmonizes disease state annotations across studies, and employs the probabilistic scVI model to construct an integrated reference anchored on healthy controls. This analysis generates a comprehensive patient sample list (131 samples from 12 independent studies) and provides harmonized cell type annotations spanning immune and non-immune populations. Based on these annotations, cell composition profiling quantifies the relative abundance of each cell type, revealing dynamic shifts in immune cell proportions during infection, with dominant populations including Tcm/Naïve helper T cells, Tem/Temra cytotoxic T cells, Tem/Effector helper T cells, and classical monocytes. At the molecular level, differential gene expression analysis delineates transcriptional programs significantly up- or downregulated in specific cell types during infection, uncovering HIV-induced immune perturbations and transcriptional remodeling. Functional pathway enrichment analysis further contextualizes these changes within biological processes and signaling pathways derived from GO, KEGG, and Reactome, providing mechanistic insights into infection-driven immune responses and inter-patient molecular heterogeneity. Together, this module enables high-resolution single-cell characterization of virus–host interactions and provides an extensible, clinically interpretable platform to support mechanistic and translational research in viral immunology.

Standardization, access, and download of data

The integration of single-cell multi-omics datasets from heterogeneous sources is often challenged by inconsistencies in experimental metadata, annotation standards, and the nomenclature of diseases or viral taxa, which can compromise data comparability and downstream analyses. To address these issues, scMOVIR implements a rigorous standardization pipeline grounded in authoritative biomedical ontologies. All viruses, diseases, tissues, and proteins were standardized using the authoritative databases such as NCBI Taxonomy [36], WHO ICD-11 [61], Uberon [62], and UniProt [63]. Based on this standardization, scMOVIR offers a user-friendly interface for efficient data exploration and retrieval. Users can explore the database either by performing targeted queries using keyword-based searches and structured filters or by navigating through the Browse interface, which organizes viruses and diseases in a hierarchical manner. Flexible download options include access to individual files as well as batch archival downloads using built-in platform tools. All database functionalities are freely accessible without login at https://pgx.zju.edu.cn/scmovir, facilitating seamless access for the research community.

Conclusion and perspectives

This study presents scMOVIR, a comprehensive single-cell multi-omics database dedicated to human viral infection and immune response. Although numerous single-cell datasets have been generated, there remains a lack of integrated multi-omics data and broad coverage across different virus types, infection states, and tissue sources. scMOVIR systematically integrates large-scale, high-quality single-cell datasets spanning transcriptomics, proteomics, and immune receptor repertoires, encompassing a wide spectrum of viral pathogens, disease states, and tissue sources. All data undergo rigorous standardization, annotation, and QC to ensure interoperability and consistency. The platform provides a comprehensive suite of interactive visualization and analytical tools, coupled with a user-friendly interface, to facilitate detailed exploration of multi-omics features and cellular and molecular heterogeneity in the context of viral infection.

scMOVIR will be continuously updated to incorporate emerging single-cell datasets and novel multi-omics modalities. Future expansions will encompass additional viral species, diverse clinical sample cohorts, and advanced data types such as single-cell ATAC-seq and spatial multi-omics. Furthermore, systematic integration of perturbation datasets collected before and after viral infection will establish a valuable resource for elucidating causal mechanisms of virus–host interactions and for training predictive computational models. With the rapid advancement of artificial intelligence, scMOVIR is also envisioned to interface with large language model-based frameworks, enabling the development of intelligent agents capable of automated data interpretation, hypothesis generation, and interactive knowledge retrieval. Collectively, scMOVIR provides a standardized, scalable, and openly accessible platform, offering critical support for systematic characterization of virus–host immune interactions at single-cell resolution.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by the Noncommunicable Chronic Diseases—National Science and Technology Major Project (2024ZD0524900), the Zhejiang Provincial Natural Science Foundation of China (LDT23H1901), the “Pioneer” and “Leading Goose” R&D Program of Zhejiang (2024C03003, 2025C02068), the Key Project under the Zhejiang Provincial and Ministry Co-construction Plan (WK-ZJ-2425), the National Natural Science Foundation of China (92369107, 32370712), the International Cooperation and Exchange Programs of the National Natural Science Foundation of China (W2412017, 2024VTJP1003), and the Fundamental Research Funds for the Central Universities (226-2025-00065). Funding to pay the Open Access publication charges for this article was provided by Noncommunicable Chronic Diseases—National Science and Technology Major Project (2024ZD0524900).

Data availability

All data can be viewed, accessed, and downloaded from the scMOVIR, which is freely accessible without any login requirements by all users at https://pgx.zju.edu.cn/scmovir.