Jiayu Wen

Alternative polyadenylation (APA) changes which regulatory sequences an RNA retains, creating a layer of cellular variation that gene-level expression profiles cannot resolve. Here we present scPolyA, a unified, AI-enabled framework that transforms conventional single-cell and spatial 3′ -end transcriptomes into quantitative maps of RNA-end choice. scPolyA connects de novo poly(A)-site discovery to cell- and spot-resolved measurements of terminal-site usage, 3′ untranslated region length and intronic polyadenylation, enabling differential APA analysis across cell states and tissue regions. Its integrated deep-learning module, DeepCPS, improves recognition of sparsely supported poly(A) sites and predicts cleavage positions at nucleotide resolution. An integrated graphical workspace brings parameter exploration, efficient rerunning and genome browsing together, while an APA-specialised AI agent translates natural-language research questions into executed analyses, verified outputs and evidence-guided interpretation. Across diverse single-cell protocols and spatial platforms, scPolyA achieves higher reference-atlas agreement than most evaluated de novo methods, while single-cell benchmarks demonstrate competitive runtime and memory use. Applied to cutaneous squamous cell carcinoma, scPolyA reveals recurrent 3′ untranslated region shortening in tumour keratinocytes across all ten patients and corresponding patterns in tumour-enriched regions, with tissue-level contrasts shaped by cellular composition. scPolyA extends conventional transcriptomics from gene abundance to RNA-end regulation, linking transcript choices to cellular states and tissue organisation.

Background: RNA regulation programs gene expression through sequence-encoded mechanisms—including RNA structure formation, protein binding, chemical modification, and RNA–RNA targeting—whose rules span nucleotide-scale motifs and longer-range context. RNA foundation models aim to learn transferable representations from large RNA corpora, but most rely on masked language modeling (MLM), where the model loss is computed on a small subset of positions and training relies on artificially corrupted inputs that are absent at downstream inference, introducing a pretraining–downstream discrepancy. Results: We introduce RNAElectra, a single-nucleotide–resolution RNA foundation model pretrained on diverse non-coding RNAs from RNAcentral using ELECTRA-style replaced-token detection (RTD). RTD trains a discriminator with a loss defined over all input positions on realistically corrupted sequences, providing dense supervision that better aligns pretraining with sequence-to-function fine-tuning. RNAElectra combines nucleotide-resolution tokenization with an efficient attention design to capture local regulatory motifs and longer-range dependencies within a single reusable backbone. Using a unified, sequence-only fine-tuning pipeline without task-specific architectures or auxiliary inputs, RNAElectra demonstrates strong cross-task generalization across benchmarks and downstream evaluations spanning RNA structure and function, RNA–protein and RNA–RNA interactions, RNA modification landscapes, and quantitative regulatory readouts such as translation efficiency and mRNA stability, outperforming widely used RNA foundation model baselines on the majority of evaluated tasks. In addition to predictive performance, RNAElectra supports interpretability by enabling analysis of learned representations and sequence determinants underlying model predictions. Conclusion: RNAElectra establishes RTD pretraining as a practical alternative to MLM for RNA foundation modeling, coupling dense position-wise supervision with an efficient architecture to deliver broadly transferable RNA representations. This framework provides a reusable backbone for RNA regulatory prediction and supports sequence-level RNA engineering and design.

Pre-training large language models on genomic sequences is a powerful approach for learning biologically meaningful representations. Masked language modeling (MLM) methods, such as DNABERT and Nucleotide Transformer (NT), achieve strong performance but suffer from partial token supervision, pre-training/fine-tuning mismatches, and high computational costs. We introduce NucEL, the first ELECTRA-style pre-training framework for genomic foundation models, addressing these limitations. Using a discriminator to identify tokens altered by a generator, NucEL provides comprehensive token-level supervision across all sequence positions, improving efficiency over the partial supervision of MLM. Incorporating ModernBERT's hybrid local-global attention and flash attention, NucEL offers an optimized BERT architecture for genomic modeling. Unlike 6-mer tokenization, NucEL uses single-nucleotide tokens for fine-grained resolution, boosting both efficiency and interpretability. Pre-trained on the human genome, NucEL achieves state-of-the-art results on diverse downstream tasks -- regulatory element identification (e.g., promoters, enhancers), transcription factor binding prediction, open chromatin classification, and histone modification profiling -- surpassing similarly sized MLM-based models and rivaling models 25x larger, such as NT. Ablation studies highlight optimal tokenization and masking strategies for ELECTRA-style DNA pre-training. Attention analysis reveals NucEL's superior capture of biologically relevant motifs compared to NT, providing insights into hierarchical learning and regulatory element modeling. These findings demonstrate ELECTRA-style pre-training as an efficient, effective strategy for genomic representation learning with broad implications for genomic research.

The replacement of canonical histones with their variant forms enables the dynamic and context-dependent regulation of the mammalian genome. Histone variants also play key roles in various pathological processes including malignancies. Among these, the aberrant expression of the testis-specific histone variant H2A.B contributes to the pathogenesis of Hodgkin lymphoma. The multifunctionality of histone variants is regulated by their posttranslational modifications (PTMs). However, the PTMs of H2A.B and their functional implications are unknown. Here, we demonstrate that the Amino terminus of H2A.B serves as a central hub for a diverse range of gene regulatory protein-protein interactions, orchestrated by phosphorylation and arginine methylation. This includes a mechanism whereby non–chromatin-bound H2A.B associates with SWI/SNF, which limits its access to the genome. Last, we identify phosphorylated H2A.B as a previously uncharacterized marker of active RNA polymerase II transcription start sites. These findings elucidate a central role for H2A.B in genome regulation and highlight the importance of its PTMs in modulating its multifunctional roles.

BMP signaling acts as an instructive cue in various developmental processes such as tissue patterning, stem cell proliferation, and differentiation. However, it is not fully understood how this signaling pathway generates different cell-specific outputs. Here, we have identified PRDM16 as a key co-factor for BMP signaling in the mouse brain. PRDM16 contributes to a repressive role of BMP signaling on neural stem cell (NSC) proliferation. We demonstrate that PRDM16 regulates the genomic distribution of BMP pathway transcription factors, the SMAD4/pSMAD complex, preventing the activation of cell proliferation genes. When Prdm16 is lost, the SMAD complex relocates to nearby genomic regions, leading to abnormal upregulation of BMP target genes. This function of PRDM16 is also required for the specification of choroid plexus (ChP) epithelial cells. Through a single-cell resolution fluorescent in situ approach, we have observed that genes co-repressed by SMAD and PRDM16, such as Wnt7b and several cell cycle regulators, become overexpressed in Prdm16 mutant ChP. Our findings elucidate a mechanism through which SMAD4 and pSMAD1/5/8 repress gene expression. Moreover, our study suggests a regulatory circuit composed of BMP and Wnt signaling, along with PRDM16, in controlling stem cell behaviors.

Epithelial tissue fusion requires coordinated molecular events at the ends of two epithelial structures. Regulatory mechanisms controlling these events remain largely elusive. In the Drosophila reproductive system (RS), this fusion unites the gonad and the genital disc-derived tissues, into a continuous tube. This study unveils the pivotal role of Hamlet (Ham), a Drosophila PR domain containing transcription factor, in orchestrating epithelial tissue fusion in the RS. Loss of ham leads to sterility and disconnection between the testes and seminal vesicles. Systematic analysis of Ham downstream genes reveals cytoskeletal, metabolic regulators and signaling pathway components. Ham activates genes for epithelial differentiation and remodeling, while repressing genes required for tissue growth and patterning. Using multiplexed in situ hybridization, we demonstrate spatial–temporal gene expression dynamics in contacting epithelia. Key Ham downstream effectors include E-Cadherin (E-Cad), Toll (Tl), and Wnt2 signaling pathways, regulating tissue interaction and fusion. Our findings present a comprehensive gene network crucial for heterotypic epithelial tissue fusion. Mammalian Ham orthologs PRDM3 and PRDM16 are highly expressed in epithelial tissues, suggesting a conserved role across species.

BMP signalling acts as an instructive cue in various developmental processes such as tissue patterning, stem cell proliferation, and differentiation. However, it is not fully understood how this signalling pathway generates different cell-specific outputs. Here we have identified PRDM16 as a key co-factor for BMP signalling. PRDM16 contributes to a repressive role of BMP signalling on neural stem cell (NSC) proliferation. We demonstrate that PRDM16 regulates the genomic distribution of BMP pathway transcription factors, the SMAD4/pSMAD complex, preventing the activation of cell proliferation genes. When Prdm16 is lost, the SMAD complex relocates to nearby genomic regions, leading to abnormal upregulation of BMP target genes. This function of PRDM16 is also required for the specification of choroid plexus (ChP) epithelial cells. Through a single-cell resolution fluorescent in situ approach, we have observed that genes co-repressed by SMAD and PRDM16, such as Wnt7b and several cell cycle regulators, become overexpressed in Prdm16 mutant ChP. Our findings elucidate a mechanism through which SMAD4 and pSMAD1/5/8 repress gene expression. Moreover, our study suggests a regulatory circuit composed of BMP and Wnt signaling, along with PRDM16, in controlling stem cell behaviors.

Epithelial tissue fusion requires coordinated molecular events at the ends of two epithelial structures. Regulatory mechanisms controlling these events remain largely elusive. In the Drosophila reproductive system (RS), this fusion unites the gonad and the genital disc-derived tissues, into a continuous tube. This study unveils the pivotal role of Hamlet (Ham), a Drosophila PR-domain containing transcription factor, in orchestrating epithelial tissue fusion in the RS. Loss of ham leads to sterility and disconnection between the testes (TE) and seminal vesicles (SV). Systematic analysis of Ham-downstream genes reveals cytoskeletal, metabolic regulators and signaling pathway components. Ham activates genes for epithelial differentiation and remodeling, while repressing genes required for tissue growth and patterning. Using multiplexed in situ hybridization, we demonstrate spatial-temporal gene expression dynamics in contacting epithelia. Key Ham downstream effectors include E-Cadherin (E-Cad), Toll (Tl) and Wnt2 signaling pathways, regulating tissue interaction and fusion. Our findings present a comprehensive gene network crucial for heterotypic epithelial tissue fusion. Mammalian Ham orthologs PRDM3 and PRDM16 are highly expressed in epithelial tissues, suggesting a conserved role across species.