Biology subjects

Li, X.

Publications and source records attributed to Li, X..

At least 91 records · Page 5Linked to original sources

bioRxiv2026

Multiscale spatial analysis implicates chromosomal metaloops in gene patterning across the Drosophila brain

Scores of chromosome-scale loops, or metaloops, arise in the Drosophila brain, but their spatial organization and relationship to neural gene expression patterns remain unclear. Here, we used multiplexed Optical Reconstruction of Chromatin Architecture (ORCA) to examine the multiscale spatial organization of metaloops in cross-sections of 100s of larval and adult Drosophila brains. We find metaloops form preferentially in the central regions of the brain, where they nucleate the formation of metadomains, characterized by the intermingling of distal topologically associating domains (TADs). At the sub-cellular scale, metaloops tend to arise towards the nuclear center, and multiple metaloops in the same cell have a preference to form hubs (3 or more contacts). Each brain nucleus generally harbors only a few loops or hubs. An in-depth analysis of the hub centered on DIP-epsilon, a synaptic wiring gene, identified a three-way metadomain that brings together the DIP-epsilon TAD; a distal TAD carrying a paralog of DIP-epsilon, DIP-zeta; and a putative regulatory TAD, across 3 Mb. This metadomain adopts distinct conformations depending on gene expression; cells expressing DIP-epsilon or DIP-zeta show preferential interactions between the TAD carrying the corresponding gene and the putative regulatory TAD. We posit that the neuron-specific formation of different subsets of metadomains might coordinate the expression of diverse combinations of synaptic wiring genes underlying complex brain architecture.

molecular biology↗

bioRxiv2026

An M-learner approach for heterogeneous mediation analysis with high-dimensional omics mediators

Causal mediation analysis is widely used to identify biological pathways linking exposures to outcomes, but most methods assume homogeneous mediation effects across individuals. In high-dimensional omics settings, this assumption can mask important heterogeneity driven by demographic, genetic, or environmental factors. We propose the M-high-learner, a flexible framework for detecting heterogeneous mediation effects with high-dimensional mediators. The method identifies mediators with subgroup-specific indirect effects while distinguishing them from null or homogeneous signals and controlling the type I error rate. It is computationally efficient, scalable, and yields interpretable sub-types. Simulation studies show that the proposed approach achieves high power while maintaining accurate error control. Applications to the Framingham Heart Study and the Multi-Ethnic Study of Atherosclerosis reveal that the mediation role of gene expression in sexs effect on high-density lipoprotein varies across subgroups defined by body mass index and age. Our framework provides a practical tool for uncovering heterogeneous biological mechanisms in high-dimensional genomic studies. Author SummaryBiological processes linking risk factors to disease often differ across individuals, but many existing methods assume these processes are the same for everyone. This can hide important differences between groups. We developed a powerful method to identify when these pathways vary across subgroups using large-scale molecular data. Our approach detects differences in how intermediate biological factors contribute to outcomes in populations defined by characteristics such as age and body mass index. Applying our method to population studies, we found that some biological pathways operate differently across groups, suggesting that key mechanisms may be missed when differences are ignored. Our work provides a tool to better understand how disease-related processes vary across individuals, which may support more targeted and personalized approaches to health research.

bioinformatics↗

bioRxiv2026

Dissecting the TMEM132A-EGFR Dependency to Unlock Translational Therapeutic Opportunities for Pan-Solid Tumor

Solid tumors remain refractory to conventional treatments, yet cell surface proteins, by virtue of their extracellular accessibility and critical roles in tumor signaling, represent an attractive class of targets for precision-targeted therapy. Here, we report that TMEM132A is an essential and previously unrecognized pan-cancer target. TMEM132A interacts directly with EGFR and stabilizes its expression, thereby tethering EGFR at the plasma membrane and sustaining constitutive activation of lipid synthesis. Mechanistically, the TMEM132A-EGFR axis promotes lipogenesis by facilitating SREBP nuclear translocation, which in turn upregulates ACLY and ACSS2 expression to drive acetyl-CoA production and downstream lipid biosynthesis, ultimately disrupting lipid droplet homeostasis. To therapeutically target this axis, we developed a nanobody, LFNanoT132A#3, which effectively blocks the TMEM132A-EGFR interaction, abrogates downstream signaling activation, and potently inhibits proliferation across multiple solid tumor types. Notably, LFNanoT132A also exerts robust antitumor activity against H1975 xenografts, a model resistant to first- and second- generation EGFR inhibitors, underscoring its potential to overcome conventional drug resistance. Our findings establish TMEM132A#3 as a critical node in membrane-tethered oncogenic signaling and metabolic rewiring, and position LFNanoT132A#3 as a promising therapeutic candidate for precision cancer therapy.

cancer biology↗

bioRxiv2026

Near-infrared optoacoustic modulation of the blood-brain barrier permeability using size-tuned hyperbranched gold nanoconstructs

The blood-brain barrier (BBB) constitutes a major bottleneck for the systemic delivery of most therapeutic agents to the central nervous system. Here, we report near-infrared reversible optoacoustic modulation of the BBB permeability (NIR-ROAMBBB), leveraging endothelial tight junction targeting hyperbranched gold nanoconstructs (HBGNCs) to amplify localized optoacoustic transduction under femtosecond laser excitation. We first synthesized HBGNCs with tunable particle sizes (62-150 nm) and consistent branch morphologies via a seed-mediated growth approach, and uncovered a non-monotonic relationship between particle dimension and optoacoustic output, where the 62 nm HBGNCs generated nearly twofold stronger optoacoustic signal than gold nanorods and gold nanostars under matched excitations. Conjugation with BV11 antibodies against junctional adhesion molecule A increased HBGNC endothelial association and cerebral accumulation, enabling focal and fluence-dependent transient BBB opening (3-6 h) under 800 nm femtosecond pulsed laser excitation, as validated by in vitro trans-endothelial electrical resistance measurements, ex vivo Evans blue extravasation staining, and in vivo NIR imaging. Featuring deep tissue penetration of NIR light, robust optoacoustic conversion of HBGNCs, and negligible femtosecond laser-induced photothermal damage, this non-invasive strategy enables precise focal modulation of BBB permeability and potential drug delivery.

bioengineering↗

bioRxiv2026

A pretrained unified model enables cellular functional profile prediction and multi-objective virtual drug screening

Cells are characterized by molecular states, coordinated molecular interactions, regulatory programs, and responses to perturbations. Systematic mapping of these cellular functional profiles across biological contexts remains experimentally costly and fragmented. Here we present InsilicoCell, a pretrained multi-modal, multi-task model that unifies prediction of cellular functional profiles spanning molecular states, molecular interactions, and perturbation-induced responses. Built on a supervised transformer architecture and pretrained on more than 88 million measurements across seven tasks, including drug sensitivity, drug-induced gene expression, and drug-protein binding, InsilicoCell learns a shared representation that links molecular profiles to cellular phenotypes, improves performance over task-specific models, and generalizes to unseen entities, contexts, and conditions. InsilicoCell extends beyond cell line systems to patient, spatial and single-cell settings, and enables multi-objective virtual drug screening. It identifies novel candidate compounds with experimental validation, including c-Myc activity inhibitors, antifibrotic agents and stemness-inducing compounds. Together, InsilicoCell provides a scalable framework for predictive cellular biology and therapeutic discovery.

bioinformatics↗

bioRxiv2026

The MOS4-associated complex subunit MAC5A maintains meristem development by regulating transcription elongation in Arabidopsis

The precise coordination of co-transcriptional RNA processing with transcriptional elongation is essential for eukaryotic gene regulation, yet the machineries coupling these processes remain largely elusive. Here, we reveal a role for the spliceosome-associated MOS4-associated complex (MAC) in regulating plant development through the direct regulation of transcriptional elongation. We demonstrate that the MAC subunit MAC5A physically interacts with Elongator Protein 6 (ELP6), a core component of the transcriptional Elongator complex. Genetic analyses reveal that MAC5A and ELP6 function synergistically to regulate apical meristem activity. Importantly, the MAC5A-Elongator module is required for the efficient expression of the critical auxin efflux carrier PIN-FORMED1 (PIN1) by enhancing RNA Polymerase II (RNAPII) occupancy across the PIN1 locus. Furthermore, we show that this elongation-promoting function is not unique to MAC5A, as other core MAC components are similarly required for efficient transcription progression. Together, our findings uncover a splicing-elongation nexus where MAC5A likely acts as a molecular bridge between the spliceosome and the elongation polymerase. This functional coupling ensures the efficient transcription of key developmental regulators, providing a mechanistic framework for the coupling of RNA processing to transcription and suggesting a conserved principle of gene expression control across eukaryotes.

plant biology↗

bioRxiv2026

Chimeric Induced Cooperativity Opens the Design Space of Eukaryotic Gene Regulation

Predictive engineering of eukaryotic transcription is limited by the coupling of signal sensing, DNA binding, TF abundance and promoter output. Here we establish chimeric induced cooperativity (CIC), a modular architecture that separates LBD, DBD and AD functions and links them to promoters with tunable basal and maximal output. Module parameters can be recombined to predict new CIC-TF configurations and guide design before construction. Ligand-induced cooperativity reduces basal DNA occupancy while increasing induced occupancy, and effective DBDs combine low OFF-state activity with strong ON-state promoter occupancy rather than binding strength alone. Synthetic promoters independently control occupancy gain and output range. The same framework extends to repression and can be recalibrated with limited measurements in mammalian cells. In yeast, CIC-12 achieved a mean fold induction of 298-fold across 12 orthogonal sensors; an earlier CIC-10 chassis enabled model-guided optimization of an eight-gene vitamin B5 biosynthetic pathway. CIC establishes a programmable, model-guided design space for eukaryotic transcriptional control.

synthetic biology↗

bioRxiv2026

A multi-tissue epigenomic atlas links the sheep non-coding genome to domestication and complex traits

Non-coding regulatory variation drives complex traits, domestication, and evolutionary adaptation, yet the sheep genome lacks high-resolution functional annotation. Here we present SheepEpimap, a multi-tissue regulatory atlas harmonizing 516 CUT&Tag histone modifications, ATAC-seq, and RNA-seq datasets across 43 adult tissues in sheep. We annotated 2.93 million cis-regulatory elements, yielding 557,441 enhancer-gene pairs and 145,407 variants with allele-specific effects. By training a sequence-to-function deep-learning model, we decoded the base-pair syntax of chromatin accessibility, annotated transcription factor motif instances genome-wide, and constructed 12,210 tissue-specific gene regulatory networks (GRNs). Integrating this resource with multi-tissue expression quantitative trait loci, selection sweeps, and genome-wide association studies prioritized non-coding variants driving domestication and complex traits. Finally, cross-species analysis revealed that sequence-conserved, tissue-matched enhancers were significantly enriched in the heritability of complex traits and diseases in humans. In summary, SheepEpimap (https://genome.ucsc.edu/s/mengzhu/SheepEpimap) provides an open-access foundational ecosystem for sheep functional genomics, precision breeding, and comparative biology.

genetics↗

bioRxiv2026

Parasitic Vectors in Aquaculture: Neobenedenia girellae and Leeches as Potential Transmission Agents for Trypanosoma carassii spectrum and Pathogenic bacteria

Parasitic infections pose multifaceted threats to farmed fish, extending beyond direct pathogenicity to facilitate infections of bacteria, viruses, and microparasites. This synergistic interaction often leads to co-infections that significantly exacerbate disease outbreaks and mortality, presenting a severe challenge to aquaculture sustainability. Recently, a novel trypanosomiasis caused by the Trypanosoma carassii spectrum has emerged in cage-cultured Larimichthys crocea along the southeast coast of China, resulting in widespread prevalence and high mortality rates. Although this pathogen is hypothesized to originate from freshwater fish, its transmission route in marine environments has remained elusive. In this study, we investigated potential vectors and intermediate hosts of T. carassii spectrum, including leeches and monogenean in natural marine settings, and simulated transmission pathways using an established laboratory model involving T. carassii spectrum, Micropterus salmoides and the leech Poecilobdella manillensis. First, our field surveys in the coast of Ningde, Fujian Province, revealed a nearly 100% co-infection rate of T. carassii spectrum and the monogenean Neobenedenia girellae in diseased juvenile L. crocea. PCR analysis detected T. carassii spectrum traces in some N. girellae specimens, and subsequent experiments confirmed that N. girellae ingests the trypanosome while feeding on host blood. Furthermore, bacterial co-pathogens, such as Vibrio harveyi, were also detected within N. girellae. We also document two fatal leech infestations: Zeylanicobdella arugamensis in hybrid groupers (Epinephelus moara [female] X Epinephelus lanceolatus [male]) in Zhangpu, and Limnotrachelobdella okae in E. lanceolatus and E. fuscoguttatus in Raoping. These leeches tested negative for trypanosomes but carried pathogenic bacteria that co-infected the host fish; nonetheless, they are established vectors for trypanosome transmission. In a laboratory cohabitation model simulating T. carassii spectrum transmission, infected M. salmoides were housed with healthy conspecifics under three conditions: Group A (with the leech P. manillensis), Group B (no leeches), and Group C (no leeches, with physical separation between infected and healthy fish). After 14 days, blood smear microscopy and PCR analysis revealed infection rates in healthy fish of 58.33% in Group A, 40.00% in Group B, and 0% in Group C. Conclusively, T. carassii spectrum can be transmitted via leeches (with higher efficiency) and may also spread through direct contact under high-density aquaculture conditions, whereas N. girellae may act as an incidental vector, further research is warranted to clarify transmission dynamics in natural marine ecosystems. Additionally, our findings highlight the role of ectoparasites, including N. girellae and leeches, as potential reservoirs and vectors for bacterial pathogens of fish. In high-density intensive aquaculture, this vectorial capacity transforms parasites from primary pathogens into key drivers of polymicrobial disease outbreaks.

zoology↗

bioRxiv2026

CodonMamba: a foundation model for programmable mRNA coding sequence design

Although mRNA codon language models provide a generalizable framework for biological sequence design, effective CDS design requires both a learned sequence design space that captures biological constraints and context-configurable design preferences. Here we present CodonMamba, a codon language model framework for mRNA prediction and programmable CDS design. Pretrained on large-scale coding-sequence corpora, CodonMamba achieved state-of-the-art performance across a comprehensive benchmark of 12 mRNA prediction tasks, ranking first on 10 of 12 tasks. Furthermore, CodonMamba establishes a programmable CDS design framework, transforming codon optimization from a process dependent on preferences embedded during model training into an inference-time steerable generation framework. By introducing user-specified codon usage priors during generation, CodonMamba enables inference-time steering toward host- or application-specific codon preferences without retraining. In design experiments, CodonMamba enabled coordinated optimization over multiple design-relevant sequence properties and demonstrated programmable cross-host CDS retargeting through inference-time prior switching, while preserving most model-derived codon choices. Together, these results establish CodonMamba as a programmable foundation model framework for CDS design, enabling context-specific mRNA sequence engineering and highlighting a promising direction for precise mRNA design using foundation models.

bioinformatics↗

bioRxiv2026

Working memory retention by medial entorhinal cortex low-dimensional neural dynamics

The ability to retain information over tens of seconds is essential for carrying out many routine tasks. Recurrently connected circuits are thought to support memory over these timescales, yet how they maintain and represent remembered information remains unclear. We asked whether the medial entorhinal cortex (mEC), where recurrent attractor dynamics generate grid-cell firing patterns and single toroidal attractor manifolds in two-dimensional environments, also supports low-dimensional neural activity patterns during memory retention. During a working-memory task, delay-period mEC dynamics did not show key features of grid firing in simple environments. Instead, broader mEC populations, not limited to grid cells, expressed multiple sequences and recurring activity patterns. Periodically repeating delay-period sequences retained task-relevant information, including past locations and future turn directions. These findings suggest that the mEC network generates low-dimensional dynamics more broadly, with grid-cell activity representing one manifestation and more complex manifold topologies supporting task-related representations during working memory.

neuroscience↗

bioRxiv2026

Trehalose exerts cryoprotective effects on piglet testicular tissue

The cryopreservation of testicular tissue is crucial for maintaining male fertility; However, its efficacy is often compromised by oxidative stress and mitochondrial dysfunction. Trehalose, a natural cryoprotectant, demonstrates significant potential, yet its specific mechanisms, particularly in mitochondrial regulation, remain insufficiently characterized. This study aimed to investigate the cryoprotective effects of trehalose on testicular tissue from 18-21-day-old piglets, with a focus on mitochondrial metabolism. Samples were cryopreserved via a slow-freezing protocol in a modified standard solution containing 200 mmol/L trehalose. The protective effect was evaluated by measuring testosterone synthesis, blood testis barrier (BTB) and spermatogenesis. Additionally, protective outcomes were assessed by measuring cell viability, tissue morphology, reactive oxygen species (ROS) levels, apoptosis rates, and testosterone secretion following freeze-thaw cycles. Transcriptomic sequencing and bioinformatics analyses were conducted to elucidate the underlying molecular mechanism. Cryopreservation led to reduced testosterone synthesis and secretion, decreased levels of BTB-binding proteins, and impaired spermatogenesis. Results indicated that 200 mmol/L trehalose significantly improved cell survival, decreased apoptosis and ROS levels, and enhanced testosterone secretion. 200 mmol/L trehalose partially increased the expression of StAR and CYP11A1 genes associated with testosterone synthesis while it protected the tight junction proteins Claudin-11, ZO-1 and the gap junction protein Cx43. Consequently, it exerted a reproductive protective effect by increasing the expression of key spermatogenic regulators DDX25, HMGB2, acrosomal protein DYP19L2, and sperm tail proteins AKAP4 and CFAP44. Transcriptomic profiling demonstrated that trehalose predominantly restored the transcriptional expression of genes involved in the mitochondrial electron transport chain and oxidative phosphorylation pathways, including ND2, COX2, ATP8, ATP6, ND5, ND6 and CYTB. These findings indicate that trehalose primarily protects piglet testicular tissue during cryopreservation by enhancing mitochondrial function, thereby providing a molecular basis for optimizing cryopreservation protocols.

cell biology↗

bioRxiv2026

PAM-DB: Revealing Protein Activation Mechanisms for Next-Generation Rational Drug Discovery

Current rational drug design relies predominantly on computational (CADD/AIDD) methods that model binding thermodynamics and static conformations of target proteins, primarily in their inactive states. However, the kinetic parameters that govern experimental efficacy-such as catalytic turnover and signaling potency-are determined by molecular interactions with transition states (TS), intermediate states (IS), and the entire continuum of conformations along the least free-energy activation pathway. The absence of this dynamic dimension has fundamentally limited the predictive power and success rate of conventional structure-based approaches. Here, we present a structural database that systematically maps the complete activation trajectories of pharmaceutically relevant targets, encompassing TS, IS, and all connecting conformational ensembles. This resource offers multiple strategic advantages for drug discovery: enabling rational targeting of previously "undruggable" proteins, facilitating biased agonism/antagonism design, revealing cryptic allosteric sites in inactive conformations, identifying novel transient pockets along the activation route, rationalizing the mechanisms of existing drugs, predicting mutational effects on activation barriers, and prospectively forecasting drug resistance and off-target liabilities. We demonstrate the utility of this database through representative case studies and provide implementation guidelines for integration into existing discovery pipelines. More detailed information can be found at our website: https://www.momedpamdb.com/en.

biophysics↗

bioRxiv2026

TM6SF2 binds cholesterol, interacts with apolipoprotein B, and promotes hepatic lipid secretion

A missense variant in TM6SF2 (transmembrane 6 superfamily member 2, TM6SF2E167K) is a major risk factor for steatotic liver disease1, while protecting against coronary artery disease2. TM6SF2 is a polytopic resident protein of the smooth endoplasmic reticulum (ER) and ER-Golgi intermediate compartment that promotes lipidation of hepatic ApoB-containing lipoproteins before secretion into the circulation. Here, we used cryo-electron microscopy (cryo-EM) to determine the structures of TM6SF2 and TM6SF2E167K at 3.64 [A] and 3.58 [A] resolution, respectively. TM6SF2 comprises 10 transmembrane helices that bind a single cholesterol molecule within a transmembrane cavity. The protein assembles into homodimers and homotetramers that interact with ApoB. Structural and biochemical analyses show that the E167K substitution reduces cholesterol binding and ApoB interaction without disrupting overall protein structure. Expression of wild-type, but not mutant, TM6SF2 restores hepatic triglyceride secretion in TM6SF2-deficient hepatocytes. Together, these findings establish the first structural framework for the bulk lipidation step in hepatic lipoprotein biogenesis, the principal pathway for hepatic triglyceride and cholesterol export into the circulation.

cell biology↗

bioRxiv2026

Contrasting selective pressures shape human pepsinogen A gene copy-number variation across Eurasia

Diet has repeatedly shaped human genomes, yet adaptation of protein digestion remains poorly understood. Using 1,348 haplotype-resolved assemblies, we reconstructed the structural evolution of the human pepsinogen A (PGA) locus and found a west-to-east increase in copy-number across Eurasia that tracked regional reliance on plant-derived protein. Modern and ancient genomes further revealed a recent selective sweep in the East but signatures of balancing selection in the West. We traced this divergence primarily to expansion of PGA34A, the most proteolytically active paralog in vitro, and showed that recurrent nonallelic homologous recombination continually generated structural diversity in this locus. Independent PGA expansions were also enriched in plant-dominant mammals. These findings link paralog-specific dosage variation in protein digestion to dietary adaptation across human populations and diverse mammalian lineages.

genomics↗

bioRxiv2026

WHaloForce enables chemigenetic imaging of molecular tension in living cells and animals

Piconewton forces borne by individual proteins within complex assemblies underlie cell adhesion, migration, and tissue morphogenesis, yet remain difficult to image in living systems. Here, we introduce WHaloForce, a chemigenetic HaloTag-based tension sensor that converts force-dependent relief of tryptophan-mediated dye quenching into a fluorescence lifetime change. Optical tweezers revealed a switch-like unquenching transition near 5 pN, and the sensor responded reversibly to force changes in cells. WHaloForce enabled quantitative tension imaging of diverse force-bearing proteins (vinculin, E-cadherin, -catenin, and laminin) in mammalian cells, mouse tissue, and C. elegans. Bright synthetic dyes made tension measurements possible at endogenous expression levels. In C. elegans, vinculin and laminin showed opposite tension patterns between tissues, revealing distinct force-transmission routes through adhesions and the extracellular matrix. During ovulation, laminin tension accumulated over repeated stretch-relaxation cycles, scaling with cumulative loading history. WHaloForce thus offers a modular platform for imaging spatiotemporal tension patterns in living systems.

cell biology↗

bioRxiv2026

A Centimeter-Scale, Peristaltic Human Intestinal Organoid with Integrated Neuro-Immune-Vascular Systems Recapitulates Enteritis and Orthotopic Colorectal Cancer

Human pluripotent stem cell-derived intestinal organoids hold great promise for disease modeling, drug screening, and regenerative medicine. However, conventional intestinal organoids are predominantly epithelial, small in scale, and lack the multicellular complexity required to recapitulate the pathophysiology of intestinal disorders such as inflammatory bowel disease (IBD) and colorectal cancer (CRC). Here, we report the development of Centimeter-Scale, purely 3D self-organized human intestinal organoids (IOs) from induced pluripotent stem cells (iPSCs) that encompass multiple tissue lineages, including epithelium, mesenchyme, smooth muscle, neurons, immune cells, and vasculature. These organoids achieve functional maturation by day 100+, exhibiting rhythmic peristaltic-like contractions, and by day 147 they display histological structures including lumens, crypt-like architecture, goblet cells, and smooth muscle. Importantly, for the first time, the neuro-muscle lineages arise spontaneously and autonomously in a purely 3D culture system, without any external stimulation (e.g., electrical, chemical, or mechanical), and mature to form functional neuromuscular junctions, driving macroscopically visible peristaltic-like contractions that mimic intestinal motility entirely through in vitro culture, without any xenotransplantation. Single-cell RNA sequencing at day 115 identified 12 cell subtypes across four major lineages, recapitulating the cellular diversity of the developing human intestine. Using this platform, we established an LPS/IFN-{gamma}-induced IBD model that recapitulated key pathological features, including epithelial disruption, immune cell infiltration, and IL-6 elevation. Transcriptomic analysis confirmed activation of the NF-{kappa}B and JAK2-STAT3 pathways, multi-modal cell death, and immune recruitment machinery, all consistent with clinical IBD pathology. Furthermore, we developed intestinal cancer models at 7 and 21 days showing abnormal hyperplasia, and a probiotic co-culture system demonstrating anti-inflammatory efficacy. Together, these results establish Centimeter-Scale intestinal organoids as a physiologically relevant, multicellular platform for modeling intestinal diseases and evaluating therapeutic interventions.

cell biology↗

bioRxiv2026

Automating scientific annotations for open transcriptomic profiles via multi-stage agents

Public transcriptomic repositories contain millions of samples, yet their large-scale reuse is hindered by heterogeneous and inconsistently reported metadata. In the Gene Expression Omnibus (GEO), key biological information is often distributed across study- and sample-level records, requiring context-dependent interpretation. Here we present GEOMeta, a large language model (LLM)-based multi-stage workflow with task-specialized agents for automated GEO metadata curation. The pipeline separates metadata retrieval, task-specific information extraction, field standardization, ontology mapping and quality control. Using GEOMeta, we generated standardized annotations for approximately 600,000 human bulk RNA-seq samples. To demonstrate its utility, we benchmarked transcriptome representation models for predicting sex, age, tissue and disease from transcriptome embeddings. We further prospectively annotated newly submitted GEO studies and evaluated 22 frontier LLMs. Recent open-source Flash models achieved annotation quality comparable to leading reasoning models while reducing costs by an order of magnitude. GEOMeta provides a scalable resource and reproducible framework for metadata curation.

bioinformatics↗