Biology subjects

Gao, R.

Publications and source records attributed to Gao, R..

At least 37 records · Page 2Linked to original sources

bioRxiv2026

Fusion gene discovery in single cells from high throughput long read single cell transcriptomes

High throughput long-read single cell RNA sequencing allows for fusion gene discovery in thousands of cells in parallel, which is however constrained by the lack of robust computational tools. Here, we develop LongFUSE that employs the XOR logic operation and stringent filtering criteria for the accurate detection of cell-specific fusion genes and their splicing isoforms. LongFUSE outperforms existing tools in both simulation and real data, demonstrating high accuracy and robustness.

bioinformatics↗

bioRxiv2026

Hypatia: a set of quantitative methods for profiling isoform across cell populations

High-throughput long-read single-cell RNA-sequencing enables isoform-level study across single cells, yet methods for systematically assessing cell-to-cell variations remain limited. Here, we develop Hypatia, a comprehensive platform for dissecting isoform complexities across cell populations, devising Tsallis entropy and Cramers V to facilitate robust comparative profiling. Hypatia revealed prominent isoform species variations and usage shifts across cell-types in glioblastoma, renal cell carcinoma, and heart, highlighting clinically relevant applications for studying isoform-derived, cell-specific functions.

bioinformatics↗

bioRxiv2025

Semantic reasoning takes place largely outside the language network

The brains language network is often implicated in the representation and manipulation of abstract semantic knowledge. However, this view is inconsistent with a large body of evidence suggesting that language processing is neurally distinct from the rest of cognition. Here, we use precision brain imaging to uncover a set of brain regions, separate from the language network, that are engaged in semantic processing of both linguistic and pictorial stimuli. In three fMRI experiments, participants (total n=41 tested across 49 sessions) viewed sentences and pictures depicting simple events. In separate blocks, they performed either a semantic task or a difficulty-matched perceptual task. Across all three experiments, several areas in left lateral prefrontal cortex, left temporo-parietal cortex, and right cerebellum responded to semantic tasks for both sentences and pictures. These semantic processing areas are spatially and functionally distinct from the nearby language-selective areas, as well as from the multiple demand and default mode networks, exhibiting a unique response profile. Our results provide evidence for a new kind of selectivity in the human brain and pave the way for future explorations of the neural mechanisms that underlie semantic reasoning.

neuroscience↗

bioRxiv2025

The language network responds robustly to sentences across diverse tasks

A network of left frontal and temporal brain areas supports language comprehension and production, implementing computations related to word retrieval and combinatorial linguistic processing. Here, we ask: to what extent are responses to language in this language network stable across task contexts, and how does this stability compare to task sensitivity in the domain-general multiple demand (MD) network? Participants (n=52) read sentences and nonword lists under six task conditions, including passive reading, reading with a memory probe after each stimulus, and reading and answering questions that require deep semantic engagement. The sentences>nonwords contrast isolated the same set of language-responsive voxels across all tasks; the locations of those voxels were participant-specific, highlighting the value of individual-specific functional localization. We therefore conclude that language localization is robust to task variation. We then examined the magnitudes and fine-grained activation patterns in these language-responsive voxels (the language network) and in the domain-general MD network, to test whether task demands modulate linguistic computations and/or recruit a distinct brain system. The language network responded robustly to sentences across all tasks, with somewhat higher responses to semantically engaging tasks. In contrast, the MD network responded to both sentences and nonwords in the presence of a task, which warrants caution when using language paradigms that include task demands, as such paradigms engage two independent networks. A multivariate analysis further revealed that stimulus information is more easily decodable in the language network, whereas task information is more decodable in the MD network. These results suggest that the language and MD networks perform complementary functions during task-driven language comprehension, with the language network primarily extracting information from linguistic input and the MD network determining the appropriate response to the task.

neuroscience↗

bioRxiv2025

(3R, 7S)-11-hydroxy-jasmonic acid is a major oxidative shunt product of jasmonate catabolism in Arabidopsis thaliana

Jasmonoyl-L-isoleucine (JA-Ile) is a pivotal oxylipin plant hormone that regulates numerous physiological processes and stress responses. Because JA-Ile is synthesized from jasmonic acid (JA), the cellular level of JA is tightly controlled. Although the oxidative catabolism of JA to 12-hydroxy-JA (12-OH-JA) is well-characterized, the biosynthesis, accumulation dynamics, and biological functions of 11-hydroxy-JA (11-OH-JA) remain poorly understood. Here, we conducted a comprehensive investigation of 11-OH-JA in Arabidopsis thaliana, encompassing its chemical synthesis, accumulation patterns, biological activity, and biosynthetic pathways. Using chemically synthesized 11-OH-JA of naturally occurring (3R,7S)-stereochemistry, we identified 11-OH-JA as the predominant hydroxylated JA derivative that accumulates following wounding. Enzymatic assays and in silico docking studies revealed that JOX1/2/3/4 exclusively catalyze the conversion of JA to 11-OH-JA, whereas 12-OH-JA is produced by distinct pathways. Notably, the quadruple mutant joxQ completely abolished 11-OH-JA accumulation without affecting 12-OH-JA levels. Furthermore, 11-OH-JA exhibited no binding affinity for the COI1-JAZ co-receptor, indicating that it acts as a biologically inactive shunt product of JA catabolism. Our findings establish JOX-mediated 11-hydroxylation of JA as a primary inactivation pathway, separate from 12-hydroxylation. Parallel, non-redundant inactivation routes deliver layered signal termination of costly defense signals. This study provides new insights into JA catabolism, highlighting 11-OH-JA as a crucial factor in the attenuation of jasmonate signaling, and thereby redefining the framework of JA turnover in A. thaliana.

plant biology↗

bioRxiv2025

Study on Early pregnancy diagnosis in sows based on joint analysis of vaginal secretion metabolomics and machine learning

ObjectiveMetabolic transformations in female mammals throughout gestation manifest in the metabolite profiles of bodily fluids, potentially serving as biomarkers for the diagnosis of pregnancy. This investigation concentrated on sows 18 days post-mating, with the objective of identifying biomarkers pertinent to early pregnancy diagnosis through the analysis of variations in metabolic molecules within vaginal secretions. MethodsVaginal secretion samples were procured from both pregnant and non-pregnant sows 18 days post-mating, subsequently subjected to untargeted metabolomic analysis utilizing liquid chromatography-mass spectrometry (LC-MS). The samples were partitioned into training and testing datasets, and machine learning algorithms--specifically, Random Forest (RF) and support vector machine (SVM)--were integrated with metabolite weight ranking to identify key biological markers. ResultsA total of 3,249 metabolic molecules were identified, among which 534 exhibited significant differential expression in the vaginal secretions of pregnant versus non-pregnant sows. Notably, three hormones that correlate with the pregnant status of the sow were discerned: Progesterone, 3-Deoxyestradiol, and Prostaglandin E1. KEGG pathway analysis revealed that the differentially expressed metabolites were predominantly enriched in nucleotide metabolic pathways. The RF model demonstrated an impressive accuracy of 1.00 in the training dataset and 0.88 in the testing dataset, while the SVM model maintained an accuracy of 1.00 across both datasets. Utilizing model weights and differential expression data, three pivotal metabolites were identified: Indolepropionic acid, cis-Aconitate, and 4-p-Coumaroylquinic acid, which exhibited ROC curve AUC values of 0.90, 0.89, and 0.79, respectively. ConclusionThis study validates the viability of utilizing vaginal secretions for the early diagnosis of pregnancy in sows, thereby establishing a foundation for the advancement of non-invasive diagnostic methodologies.

zoology↗

bioRxiv2025

SVPG: A pangenome-based structural variant detection approach and rapid augmentation of pangenome graphs with new samples

Breakthrough advances in long-read sequencing technologies have opened unprecedented opportunities to study genetic variations through comprehensive pangenome analysis. However, the availability of structural variant (SV) calling tools that can effectively leverage pangenome information is limited. In addition, efficient construction of pangenome graphs becomes increasingly challenging with acquisition of larger number of samples. In this study, we present SVPG, an approach that leverages haplotype-resolved pangenome reference for accurate SV detection and rapid pangenome graph augmentation from long-read sequencing data. Compared to state-of-the-art SV callers, SVPG maintained superior overall performance across different sequencing technologies and coverages. SVPG also achieved notable improvements in calling rare and individual-specific SVs on both simulated and real somatic datasets. Furthermore, in a benchmark involving 20 samples, SVPG accelerated pangenome graph augmentation by nearly 10-fold compared to traditional augmentation strategies. We believe that this novel SVPG method, has the potential to revolutionize SV detection and serve as an effective and essential tool, offering new possibilities for advancing pangenomic research.

bioinformatics↗

bioRxiv2025

Features fusion or not: harnessing multiple pathological foundation models using Meta-Encoder for downstream tasks fine-tuning

The emergence of diverse pathological foundation models has empowered computational pathology tasks, including tumor classification, biomarker prediction, and RNA expression prediction. However, variations in model architecture and data sources lead to inconsistent downstream performance and complicate centralized training. Specifically, the lack of data sharing makes retraining foundation models with pooled data infeasible. Alternatively, the release of model parameters enables combining multiple models during fine-tuning. Inspired by the meta-analysis method, we propose the Meta-Encoder framework, which integrates features from multiple foundation models to generate a comprehensive representation, improving downstream fine-tuning task performance. Comparative experiments demonstrate that Meta-Encoder is more effective than individual foundation models, with its strengths more pronounced in handling complex tasks. While single models may perform sufficiently well for simple tasks, Meta-Encoder can match or even surpass the best-performing single model, alleviating concerns over model selection. Moderately challenging tasks benefit from Meta-Encoders concatenation or self-attention strategies, with the latter demonstrating superior performance in more challenging scenarios. For highly complex tasks, such as high-dimensional gene expression prediction, self-attention proves to be the most effective Meta-Encoder strategy, balancing feature integration and computational efficiency. For three patch-level spatial gene expression prediction tasks (HEST-Benchmark, CRC-inhouse, and Her2ST), the self-attention strategy improved the Pearson correlation by 38.58%, 26.06%, and 20.39%, respectively, compared to the average performance of three patch-level single models. Similarly, for the TCGA-BRCA, TCGA-NSCLC, and TCGA-CRC WSI-level bulk gene expression prediction tasks, the Pearson correlation increased by 14.36%, 9.27%, and 42.55%, respectively, compared to the average performance of two WSI-level single models. By leveraging multiple pathological foundation models using Meta-Encoder, it can further improve molecular characterization in pathology images to advance precision oncology.

bioinformatics↗

bioRxiv2025

Sequence based prediction of cell type specific microRNA binding and mRNA degradation for therapeutic discovery

MicroRNAs and RNA binding proteins are crucial elements of post-transcriptional gene regulation, which governs the fate of mRNA molecules in the cell. However, the landscape of these regulatory interactions, particularly across different mammalian cell types, remains underexplored. We describe REPRESS, a deep learning model that predicts cell-type-specific microRNA binding and mRNA degradation directly from RNA sequence. REPRESS was trained on AGO2-CLIP, miR-eCLIP and Degradome-Seq data profiling millions of microRNA binding and mRNA degradation sites across multiple cell types in human and mouse. It reveals biology that other state-of-the-art methods did not, such as identifying repressive non-canonical miRNA target sites and decoding the regulatory effects of sequence context and miRNA binding site multiplicity. REPRESS outperforms other advanced methods and neural architectures on a comprehensive suite of seven orthogonal tasks, including identifying genetic variants that affect microRNA binding, predicting out-of-distribution data from massively parallel reporter assays, and predicting canonical and non-canonical miRNA mediated repression. To demonstrate the general utility of REPRESS, we show that it provides insights into novel biology and the design of RNA therapeutics. Code is available at : https://github.com/deepgenomics/repress

genomics↗

bioRxiv2025

Transcriptome and metabolome analyses reveal novel genetic targets for L-tryptophan overproduction in Corynebacterium glutamicum

Corynebacterium glutamicum is a promising microbial chassis for the industrial production of L-tryptophan, which has exhibited increasing demand due to its diverse applications and high market value. In previous work, we developed an L-tryptophan-overproducing C. glutamicum strain TR26 through multiple rounds of rational metabolic engineering. Here, comparative transcriptome and metabolome analyses were conducted between TR26 and its progenitor strain MB001 to reveal the underlying mechanisms and potential bottlenecks for L-tryptophan production in TR26. Furthermore, by systematically down- and up-regulating differentially expressed genes of interest, two novel genetic targets, glnK and sugR, were identified as being associated with L-tryptophan synthesis. Specifically, the repression of glnK and overexpression of sugR in strain TR26 increased the titer of L-tryptophan by 6.7% and 20.9%, respectively. Gene transcription profiling and intracellular metabolite analysis further suggested that the observed improvements in L-tryptophan synthesis could be attributed to optimized nitrogen transport and metabolism, efficient reallocation of cellular resources and enhanced supply of phosphoenolpyruvate (PEP). This study advances our understanding of the regulation mechanisms governing L-tryptophan synthesis in C. glutamicum and provides valuable insights for further optimization of industrial cell factories.

synthetic biology↗

bioRxiv2025

Reference-guided genome assembly at scale using ultra-low-coverage high-fidelity long-reads with HiFiCCL

Population genomics using short-read resequencing captures single nucleotide polymorphisms and small insertions and deletions but struggles with structural variants (SVs), leading to a loss of heritability in genome-wide association studies. In recent years, long-read sequencing has improved pangenome construction for key eukaryotic species, addressing this issue to some extent. Sufficient-coverage high-fidelity (HiFi) data for population genomics is often prohibitively expensive, limiting its use in large-scale populations and broader eukaryotic species and creating an urgent need for robust ultra-low coverage assemblies. However, current assemblers underperform in such conditions. To address this, we propose HiFiCCL, the first assembly framework specifically designed for ultra-low-coverage high-fidelity reads, using a reference-guided, chromosome-by-chromosome assembly approach. We demonstrate that HiFiCCL improves ultra-low-coverage assembly performance of existing assemblers and outperforms the state-of-the-art assemblers on human and plant datasets. Tested on 45 human datasets ([~]5x coverage), HiFiCCL combined with hifiasm reduces the length of misassembled contigs relative to hifiasm by an average of 21.19% and up to 38.58%. These improved assemblies enhance germline structural variant detection, reduce chromosome-level mis-scaffolding, enable more accurate pangenome graph construction, and improve the detection of rare and somatic structural variants based on the pangenome graph under ultra-low-coverage conditions.

bioinformatics↗

bioRxiv2025

Phase separation of Polycomb-like (PCL) proteins drive PRC2 complex condensates to regulate gene expression

Polycomb group (PcG) proteins function as two major multicomponent protein complexes: Polycomb repressive complex 1 (PRC1) and 2 (PRC2) to repress key developmental genes and maintain epigenetic memory of cell identity during development. Phase-separation has long been implicated in PRC1 mediated gene silencing, but it is still unclear whether PRC2 also utilize a similar mechanism to regulate gene expression. Here we report that Polycomb-like (PCL) proteins, PRC2 accessory factors, can phase-separate in vitro and generate dynamic puncta in vivo. Biochemical and cellular analyses reveal that PCL proteins (hereafter referred to as PCLs) have intramolecular interaction between N- and C-terminal domains, which make PCLs into more compact conformations. The intramolecular interaction not only controls the size of the PCLs phase separation droplets, but also affects the chromatin association of PRC2. Finally, we show that CpG islands, key DNA regulatory elements in mammalian promoters, disrupt the N-C intramolecular interaction of PCLs to expose their middle intrinsically disordered regions (IDRs), which in turn trigger PCLs driven PRC2 condensates formation. Together, this study provides a new perspective on the regulation of PRC2 by PCLs, implicating PCLs read CpG islands to fine-tune their oligomerization to overcome the threshold for PRC2 recruitment to corresponding genomic loci via multivalent interaction with CpG islands chromatin.

biochemistry↗

bioRxiv2025

Homoharringtonine Promotes FTO Degradation to Suppress LILRB4-Mediated Immune Evasion in Acute Monocytic Leukemia

Acute monocytic leukemia (AML-M5), a subtype of acute myeloid leukemia, is a highly aggressive malignancy characterized by a poor prognosis, primarily due to the ability of leukemic cells to evade immune surveillance. In this study, we demonstrate that homoharringtonine (HHT), an FDA-approved therapeutic agent for chronic myeloid leukemia (CML), inhibits this immune evasion by targeting the FTO/m6A/LILRB4 signaling pathway in monocytic AML. Utilizing RNA sequencing (RNA-seq) and various functional assays, we reveal that HHT treatment significantly reduces LILRB4 expression at both the RNA and protein levels, suggesting that the effects of HHT on LILRB4 are distinct from its well-established role as a protein synthesis inhibitor. Mechanistically, HHT treatment markedly increases global levels of RNA m6A in THP-1 cells by promoting the degradation of FTO, which subsequently diminishes the expression of its downstream targets, MLL1 and LILRB4. Furthermore, in vitro and in vivo analyses employing monocytic AML cell lines, mouse-derived AML xenograft models, and patient samples collectively support the conclusion that HHT suppresses immune evasion in monocytic AML by reducing LILRB4 expression. Importantly, the downregulation of LILRB4 resulting from HHT treatment enhances the susceptibility of THP-1 cells to CD8+ T cell cytotoxicity, accompanied by increased markers of immune activation. Overall, our findings position HHT as a promising clinical agent for enhancing CD8+ T cell-based cancer immunotherapy by mitigating immune evasion in monocytic AML. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=147 HEIGHT=200 SRC="FIGDIR/small/643466v1_ufig1.gif" ALT="Figure 1"> View larger version (20K): org.highwire.dtl.DTLVardef@e72e5dorg.highwire.dtl.DTLVardef@19519aeorg.highwire.dtl.DTLVardef@1c8f786org.highwire.dtl.DTLVardef@f2a963_HPS_FORMAT_FIGEXP M_FIG C_FIG

cancer biology↗

bioRxiv2025

High-Parameter Spatial Multi-Omics through Histology-Anchored Integration

Spatial omics face challenges in achieving high-parameter, multi-omics co-profiling. Serial-section profiling of complementary panels mitigates technical trade-offs but introduces the spatial diagonal integration problem. To address this, we present SpatialEx and its extension SpatialEx+, computational frameworks leveraging histology as a universal anchor to integrate spatial molecular data across tissue sections. SpatialEx combines a pre-trained H&E foundation model with hypergraph and contrastive learning to predict single-cell omics from histology, encoding multi-neighborhood spatial dependencies and global tissue context. SpatialEx+ further introduces an omics cycle module that encourages cross-omics consistency via slice-invariant mappings, enabling seamless integration without co-measured training data. Extensive validations show superior H&E-to-omics prediction, panel diagonal integration, and omics diagonal integration across various biological scenarios. The frameworks scale to datasets exceeding one million cells, maintain robustness with non-overlapping or heterogeneous sections, and support unlimited omics layers in principle. Our work makes multi-modal spatial profiling broadly accessible.

bioinformatics↗

bioRxiv2025

Size switchable DNA origami structure enabled by dynamic crossovers

DNA nanostructures are formed by many DNA double helices connected via crossovers. If multiple DNA double helices are connected by crossovers at different base positions (different rotational angles), these DNA double helices are stacked together to obtain desired two- or three-dimensional DNA structures. Here we inserted i-motif or hairpin motifs in crossover regions of one layer DNA origami structure and employed dynamic DNA element to implement size-switchablity between expansion and contraction. By changing pH or adding complementary strands to regulate interspacings between DNA double helixes, multiple crossover replacements realize the dynamic size changes between 68 nm and 150 nm as characterized by atomic force microscopy. We further demonstrated dynamic sizes changes enabled or inhibited FRET (fluorescence resonance energy transfer) signals between Cy3 and Cy5 and the controlled distances between protein molecules. The results reveal the feasibility of fabricating size-switchable DNA origami nanostructures via crossover redesign, demonstrating their potential in nano-engineering, proximity enabled chemical reacitons, as well as for biomedical applications.

bioengineering↗

bioRxiv2025

A brainstem map of orofacial rhythms

Rhythmic orofacial movements, such as eating, drinking, or vocalization, are controlled by distinct premotor oscillator networks in the brainstem. Orofacial movements must be coordinated with rhythmic breathing to avoid aspiration and because they share muscles. Understanding how brainstem circuits coordinate rhythmic motor programs requires neurophysiological measurements in behaving animals. We used Neuropixels probe recordings to map brainstem neural activity related to breathing, licking, and swallowing in mice drinking water. Breathing and licking rhythms were tightly coordinated and phase-locked, whereas intermittent swallowing paused breathing and licking. Multiple clusters of neurons, each recruited during different orofacial rhythms, delineated a lingual premotor network in the intermediate nucleus of the reticular formation (IRN). Local optogenetic perturbation experiments identified a region in the IRN where constant stimulation can drive sustained rhythmic licking, consistent with a central pattern generator for licking. Stimulation to artificially induce licking showed that coupled brainstem oscillators autonomously coordinated licking and breathing. The brainstem oscillators were further patterned by descending inputs at moments of licking initiation. Our results reveal the logic governing interactions of orofacial rhythms during behavior and outline their neural circuit dynamics, providing a model for dissecting multi-oscillator systems controlling rhythmic motor programs.

neuroscience↗

bioRxiv2024

Bridging the Dimensional Gap from Planar Spatial Transcriptomics to 3D Cell Atlases

Spatial transcriptomics (ST) has revolutionized our understanding of tissue architecture, yet constructing comprehensive three-dimensional (3D) cell atlases remains challenging due to technical limitations and high cost. Current approaches typically capture only sparsely sampled two-dimensional sections, leaving substantial gaps that limit our understanding of continuous organ organization. Here, we present SpatialZ, a computational framework that bridges these gaps by generating virtual slices between experimentally measured sections, enabling the construction of dense 3D cell atlases from planar ST data. SpatialZ is designed to operate at single-cell resolution and function independently of gene coverage limitations inherent to specific spatial technologies. Comprehensive validation using real 3D ST and independent serial sectioning datasets demonstrates that SpatialZ accurately reconstructs virtual slices while preserving cell identities, gene expression patterns, and spatial relationships. Leveraging the BRAIN Initiative Cell Census Network data, we constructed a 3D hemisphere atlas comprising over 38 million cells, a scale not feasible experimentally. This dense atlas enables unprecedented capabilities, including in silico sectioning at arbitrary angles, explorations of gene expression across both 3D volumes and surfaces, and 3D mapping of query tissue sections. While currently validated for spatial transcriptomics, the underlying principles of SpatialZ could potentially be adapted for spatial proteomics, spatial metabolomics, and even spatial multi-omics. Validated through internal and external testing, our computationally generated atlas maintains biological accuracy, providing unprecedented resolution of spatial molecular landscapes and demonstrating the potential of computational approaches in advancing 3D ST.

bioinformatics↗

bioRxiv2024

NicheTrans: Spatial-aware Cross-omics Translation

Spatial omics technologies have revolutionized our studies on tissue architecture and cellular interactions at single-cell resolution. While spatial multi-omics approaches offer unprecedented insights into complex biological systems, their widespread adoption is hindered by technical challenges, specialized requirements, and limited accessibility. To address these limitations, we present NicheTrans, the first spatially-aware cross-omics translation method and a flexible Transformer-based multi-modal deep learning framework. Unlike existing single-cell (non-spatial) translation methods, NicheTrans uniquely incorporates both cellular microenvironment information and flexible integration of multi-modal data, such as morphology and prior knowledge. We validated NicheTrans across diverse biological cases: Parkinsons Disease (PD), Alzheimers Disease (AD), breast cancer, and lymph nodes. Our approach demonstrated superior performance compared to existing single-cell methods, highlighting the crucial role of spatial and multi-modal information in cross-omics translation. Through NicheTrans, we uncovered spatial multi-omics domains that were not detectable through single-omics analysis alone. Model interpretation revealed key molecular relationships, including gene programs associated with dopamine metabolism and amyloid {beta}-associated cell states. Additionally, using translated protein markers as spatial landmarks, we quantified the spatial organization of key glial cell subtypes in the AD brain. NicheTrans represents a powerful tool for generating comprehensive spatial multi-omics insights from more accessible single-omics measurements, making multi-omics analysis more feasible for the broader research community.

bioinformatics↗