Shengquan Chen

Single‐cell DNA methylation (scDNAm) sequencing provides unique insights into epigenetic heterogeneity and cell‐specific regulatory landscapes. However, accurate cell type annotation for scDNAm data remains challenging, as the distinct data distribution of scDNAm hinders the adaptation of annotation methods from other omics, and specialized annotation tools for scDNAm are currently lacking. Here, MethyAnno is proposed as an interpretable deep metric learning framework that leverages multi‐scale information for accurate cell type annotation of scDNAm data. Additionally, MethyAnno enables generalized category discovery in open‐set scenarios by utilizing density‐based clustering to automatically estimate the number of novel cell types, while simultaneously deciphering cell‐type‐specific epigenetic signatures for biological interpretability. Extensive experiments demonstrate that MethyAnno excels in cross‐dataset annotation and novel type discovery, showing exceptional robustness in few‐shot scenarios for rare cell types. Moreover, interpretability analysis in the human brain dataset correctly recovers the genetic link between Sst interneurons and epilepsy heritability, the association of OPC cells with Alzheimer's disease, as well as the regulatory role of Pvalb cells in synaptic plasticity. Taken together, these findings establish MethyAnno as a robust and biologically interpretable tool for accurate cell type annotation and downstream epigenetic analysis.

Accurate cell-type annotation is essential for revealing the dynamic, cell-type-specific accessibility of regulatory elements from single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) data. However, unlike the more mature single-cell RNA-sequencing (scRNA-seq) cell-type annotation workflows, scATAC-seq cell-type annotation remains challenging owing to extreme sparsity, high dimensionality, the scarcity of labeled scATAC references, and pronounced batch effects across data sets. To enable annotation without relying on extensive scATAC labels, we introduce CARA, a cross-modal Bayesian framework that transfers cell-type knowledge from scRNA-seq to scATAC-seq. CARA simultaneously enables cell-type annotation, batch correction, and detection of cell types absent from the reference by combining pretraining on scRNA-seq data and semisupervised learning on target scATAC-seq data, along with distribution alignment, dynamic class weighting, and novel cell-type detection. Across diverse benchmark data sets, CARA consistently outperforms baseline methods, remaining robust to batch effects. CARA's cross-modal alignment and robust annotation strategy preserve fine-grained lineage structure, enabling reconstruction of the hematopoietic differentiation trajectory. Through multidimensional uncertainty and latent-space clustering, CARA identifies novel, rare, or aberrant populations outside the reference cell-type space, providing candidates for further biological validation and perturbation. Using an omic-specific generative framework and distribution alignment, CARA delivers accurate knowledge transfer and detects novel cell types in single-cell DNA methylation data, demonstrating generalization capability to gene-centric modalities. Ultimately, CARA offers a powerful and flexible solution for cross-modal cell-type annotation in complex single-cell settings, facilitating the discovery of novel cell types and mechanistic insight into cell-type-specific regulation across diverse analyses.

Single-cell DNA methylation (scDNAm) profiling is revolutionizing our understanding of epigenetic control of gene expression, but its accurate analysis is severely hindered by extreme data sparsity. While imputation methods have undergone remarkable development in recent years, a rigorous benchmark to guide method selection remains absent. We established the first systematic benchmarking framework for scDNAm imputation, subjecting five state-of-the-art methods to a comprehensive evaluation across 13 published experimental scDNAm datasets. Performance was systematically assessed across seven critical dimensions: accuracy, sensitivity to data characteristics, scalability, robustness to data splitting strategies, inter-dataset generalizability, convergence behavior, and computational efficiency. Through rigorous statistical analysis, we dissected the influence of intrinsic data attributes and model architectures on the fidelity of scDNAm imputation to provide guidance for selecting appropriate methods for given scenarios. Furthermore, based on the benchmark-identified limitations, we proposed a dual-view strategy to address the performance bottlenecks of existing methods: at the model view, we developed BridgeCpG, an ensemble strategy to integrate complementary modeling strengths to overcome single-model limitations; at the data view, we introduced an adaptive divide-and-conquer strategy to partition highly heterogeneous datasets into several homogeneous subsets amenable to accurate imputation, followed by aggregating the sub-results. This integrated framework, spanning both model and data views, delivers quantitative analyses, scenario-aware selection guidelines, and targeted innovative strategies, establishing a rigorous, enabling foundation for accurate, high-throughput, and scalable next-generation single-cell epigenomic analysis.

Single-cell DNA methylation (scDNAm) sequencing offers unprecedented opportunities for dissecting epigenetic heterogeneity and gene regulatory mechanisms. However, most existing analytical methods are adapted from conventional single-cell sequencing frameworks, such as single-cell RNA sequencing, and fail to account for the unique characteristics of scDNAm data, including the widespread presence of explicit missing values and their distinct distribution patterns. To enable systematic exploration of DNA methylation-derived epigenetic variation at single-cell resolution, we propose scMethCraft, a versatile analytical toolkit. scMethCraft integrates multi-perspective genomic sequence features and position information of methylated regions via a hybrid neural network, and iteratively quantifies the cell-to-cell associations, enabling accurate modeling of scDNAm data. scMethCraft facilitates accurate reconstruction of DNA methylation landscapes and supports a spectrum of downstream analyses, including cell embedding, multi-source data integration, cell type annotation, epigenetic signal enhancement, and identification of differentially methylated regions, thereby significantly facilitating the identification and characterization of scDNAm heterogeneity. The integration of scMethCraft with complementary analytical workflows, including biological function enrichment, tissue-specific expression analysis, and partitioned heritability assessment, enables the discovery of biological insights into specific cellular subpopulations. Furthermore, scMethCraft identified oligodendrocyte-associated genes that are not well characterized in existing databases, providing an epigenetic perspective for investigating underlying biological mechanisms.

Spatially resolved sequencing technologies have revolutionized our understanding of biological regulatory processes within tissue microenvironments by simultaneously capturing the states of genomic regions, genes, and proteins alongside the spatial organization of cells. However, inherent heterogeneity across modalities and samples poses substantial challenges for the integrative analysis of spatial omics data, underscoring the urgent need for advanced computational methods. In this study, we propose PRESENT, a contrastive learning-based integrative framework for cross-modality representation of spatial multi-omics data. PRESENT employs omics-specific encoders consisting of graph attention networks and Bayesian neural networks coupled with distribution-aware decoders to model distinct modalities, and an inter-omics alignment module for multi-omics integration. By effectively incorporating spatial dependencies with multi-omics information across diverse species and technologies, PRESENT facilitates the accurate identification of spatial domains and the elucidation of underlying regulatory mechanisms. Furthermore, PRESENT can be extended to multi-sample integration via a two-stage training workflow, which incorporates inter-batch alignment loss, intra-batch preserving loss, batch-adversarial learning, and cyclic graph refinement strategies to eliminate batch effects while retaining biological signals. Extensive experiments on tissue samples across different anatomical regions and developmental stages demonstrate that PRESENT enables the characterization of hierarchical tissue structures from a spatiotemporal perspective.

Single-cell multiomic sequencing technologies have offered unprecedented insights into cellular heterogeneity by jointly profiling gene expression and epigenetic landscapes at single-cell resolution. However, the application of these technologies remains limited owing to technical challenges and high costs. Computational approaches for cross-modality translation provide a promising solution to these limitations by enabling the inference of one modality from another. However, existing methods for cross-modality translation between single-cell RNA sequencing (scRNA-seq) and single-cell DNA methylation (scDNAm) data face limitations, including unidirectionality, inadequate modeling of context-specific DNA methylation–expression associations, neglect of biological relevance in evaluation, and poor performance in limited paired training data. To fill these gaps, we introduce scBOND, a bidirectional cross-modal translation framework tailored for scRNA-seq and scDNAm profiles. scBOND leverages a mixture-of-experts block to capture context-dependent regulatory patterns, while implementing self-attention mechanism and a feature recalibration module to enhance biological signal fidelity. Extensive experiments demonstrate scBOND consistently outperforms baseline methods in both translation directions, yielding high-accuracy translation while preserving cellular structure. In mouse embryonic data, scBOND preserves subtle, functionally significant differences between closely related cell types, which are undetected in the original data. Downstream analyses confirm that scBOND effectively recovers tissue-specific signals in human brain neurons. Moreover, using RNA-only data, we reconstruct scDNAm profiles and identify cell-type- and stage-specific regulatory mechanisms in oligodendrocyte lineage. To further improve model generalization in paired data-scarce scenarios, we propose scBOND-Aug, a variant of scBOND equipped with a biologically informed data augmentation strategy, which demonstrates superior results with limited paired data.

The rapid advancement of transposase-accessible chromatin using sequencing (ATAC-seq) technology, particularly with the emergence of single-cell ATAC-seq (scATAC-seq), accelerates the studies of gene regulation. However, the absence of a generic feature reference for ATAC-seq data limits single-cell analyses and hinders the development of comprehensive cell atlases. To address this, we construct a generic chromatin accessibility reference by aggregating peaks from 624 high-quality bulk ATAC-seq datasets, defining about 1.4 million consensus peaks (cPeaks). Leveraging a deep neural network model, we expand cPeaks to include previously unobserved regions, enhancing their coverage across diverse tissues and cell types. cPeaks exhibit consistent shapes across tissue types, sequencing technologies, and peak-calling methods, indicating that they represent inherent genomic features. Compared to existing feature-defining methods and references, cPeaks show superior performance in scATAC-seq analyses, improving cell annotation and rare cell type identification. Additionally, cPeaks provide insights into chromatin dynamics during cellular differentiation and tumor progression. cPeaks can serve as a robust reference for chromatin accessibility studies to promote cross-dataset consistency and accelerate biological discoveries.

Background: Differential gene selection is fundamental to transcriptomics; however, mainstream methods typically fit each gene independently via marginal analyses. Although multifactor designs can be accommodated, gene–gene interactions are not explicitly modeled, rendering results susceptible to systematic bias driven by coexpression patterns. Additionally, heterogeneous tools, inconsistent metrics, and fragmented workflows compromise the reproducibility and interpretability of downstream enrichment and visualization analyses. Findings: We developed TransPro, an open-source integrated framework comprising 2 complementary packages, TransProPy and TransProR, for systematic benchmarking, bias correction, and reproducible visualization in differential gene selection. TransProPy (Python) combines multivariate AUC-based complementarity quantification with ensemble learning for interaction-aware gene selection, whereas TransProR (R) provides standardized differential analysis, pathway enrichment, and 7 visualization workflows (circular dendrograms, chord diagrams, spiral plots, and interaction networks). Cross-dataset generalizability was assessed using 12 independent datasets spanning multiple cancer types and normal tissues with broad heterogeneity in data origin, platform, batch structure, and sample size. Under stringent thresholds, positively and negatively correlated genes maintained a near-equal proportion at the gene level, and activated and suppressed pathway proportions exhibited good concordance with gene-level correlation patterns; critically, TransProPy produced meaningful enrichment results under conditions where conventional methods failed. Quantitative comparisons of core-enriched gene proportions further supported these differences (Kruskal–Wallis and pairwise Wilcoxon tests, P < 0.001). Conclusions: TransPro establishes a unified, reproducible framework that corrects method-specific biases while bridging computational discovery and biological interpretation. All code, workflows, documentation, and example data are openly accessible to support reproducibility and community reuse.