Biology subjects

Zhang, C.

Publications and source records attributed to Zhang, C..

At least 37 records · Page 2Linked to original sources

bioRxiv2026

CryoMV: Structure-Prior-Guided Modeling and Real-Particle Validation of Continuous Conformational Transitions in Cryo-EM

Continuous protein conformations are essential for understanding fundamental biological processes and supporting drug discovery. Although cryo-EM can resolve individual states at high resolution, recovering continuous heterogeneity from 2D particle images remains challenging. High noise, motion blur, and limited structural priors make it difficult to accurately generate and validate high-resolution continuous conformations using raw particle data. Here, we introduce cryoMV, a framework that integrates structure-prior-guided modeling with real-particle validation for continuous conformational transitions. CryoMV uses reference density maps to establish structural anchors and motion priors, models candidate transition paths between selected conformations, and transfers the learned representation to raw 2D cryo-EM particle images. Each candidate conformation is subsequently evaluated using the estimated particle poses and contrast transfer functions. Supported conformations are reconstructed through raw particle back-projection and assessed using canonical half-maps and Fourier shell correlation. On EMPIAR-10516 and EMPIAR-10345, cryoMV achieves excellent performance in terms of robustness, verifiability, and reconstruction resolution. By incorporating structure-prior modeling and evidence from the raw particles, cryoMV offers an explicit mechanism for assessing whether generated conformations are supported by experimental data and provides a practical approach to reducing model-induced artifacts in continuous cryo-EM heterogeneity analysis.

bioinformatics↗

bioRxiv2026

PROXINAUT: An Automation Platform for Rapid Hit-to-Degrader Discovery

Massive library-based platforms enable rapid early hit discovery, yet subsequent hit resynthesis and validation remain major bottlenecks in drug discovery. This delay is amplified when optimizing hits, or elaborating structures towards more advanced modalities, such as chemical inducers of proximity (CIPs). Here, we report PROXINAUT, an end-to-end workflow integrating self-encoded library-based hit discovery with an automated parallel solid-phase synthesis platform to streamline hit discovery, validation, optimization, and degrader development. As a demonstration, screening a 176k-member benzimidazole library against BRD4 yielded six hit candidates, which were obtained via parallel automated synthesis and validated as submicromolar binders. The automation workflow empowered a rapid structure activity relationship exploration around the whole scaffold and resulted in a variant with six fold improved binding affinity. Next, we aimed to automatize the development of bifunctional degraders. Solid phase synthesis typically leaves a C-terminal amide as a synthetic artifact that can compromise druglike properties. We developed a strategy to repurpose this resin-attachment site, converting it into functional degrons featuring either FBXO31-targeting C-terminal amides or cereblon-binding cyclic imides. Operating without human intervention, our automated platform executes up to 16 synthetic steps across multiple parallel structures, enabling true de novo synthesis of full bifunctional scaffolds, where ligand, linker and E3 recruiter variations can all be explored within the same workflow. This resulted in cell-active BRD4 degraders with subnanomolar to nanomolar potency. Overall, we show how library selections can merge with multi-step parallel automation to convert hits directly into validated leads and advanced degrader modalities.

bioengineering↗

bioRxiv2026

Mechanistic Investigation of Reversible Hibernation-Driven Immune Suppression in Thirteen-Lined Ground Squirrels

Torpor in mammals imposes extreme energetic constraints, yet how it reshapes the immune system remains poorly understood. Here we combined single-cell RNA sequencing with quantitative image analysis of immune cell populations in splenic tissues to define immune remodeling during torpor in a natural hibernator, 13-lined ground squirrels. The torpid spleen showed a significant contraction of white pulp with a preferential reduction of B cell- and T cell-rich adaptive niches and relative preservation of innate myeloid populations. Across immune cell lineages, a conserved transcriptional program of metabolic downscaling emerged, marked by suppression of glycolysis, cell-cycle progression, RNA processing, and translation, together with reduced glucose transporter expression and induction of cold-shock RNA-binding proteins. Despite this pervasive metabolic quiescence, immune cells maintain lineage identity, with B cells undergoing numerical and transcriptional contraction linked to follicular remodeling and T cells adopting a quiescent, stress-resistant state without apoptotic enrichment. Innate cell populations remained numerically enriched but transcriptionally restrained, consistent with low-energy tissue surveillance rather than inflammatory activation. These findings identified torpor as a coordinated, multi-tiered state of reversible immune suppression, in which hierarchical metabolic and lineage-specific programs conserve energy while preserving essential immune infrastructure for rapid restoration. Our identified key molecules and pathways in immune suppression at the cellular and transcriptional levels have important implications for future development of targeted treatment for autoimmune diseases, cancers, and infectious diseases.

immunology↗

bioRxiv2026

Newly identified COPD GWAS protein interactors reveal a potential disease network module

Background: Chronic obstructive pulmonary disease (COPD) is a complex, heterogeneous disease influenced by genetic and environmental factors. However, the interactions between COPD risk genes and their collective role in COPD susceptibility remain largely elusive. We hypothesize that the protein-protein interaction (PPI) network of genes associated with COPD risk can provide molecular insight into the mechanisms of COPD pathogenesis. Methods: We used affinity purification mass spectrometry (AP-MS) to detect protein-protein interactions of six well-established COPD GWAS gene products (AGER, FAM13A, FBXO38, HHIP, IREB2, MFAP2) in two relevant lung cell lines (IMR90, 16HBE). These genes are located in significant COPD GWAS regions, and their physiological role in COPD has been previously confirmed by functional studies. We analyzed the impact of the newly identified interactions in contextual, cell type-specific PPI networks built from the combination of publicly available PPI data (HUBRIS) and cell type-specific gene expression data. Results: Using AP-MS, we identified 482 unique interactors across 701 newly identified protein interactions with the six GWAS gene products in at least one of the cell lines (AGER: 122, FAM13A: 83, FBXO38: 71, HHIP: 108, IREB2: 8, MFAP2: 309). Subsequent network analysis incorporating these new PPIs with public PPI databases revealed: (1) the new interactions significantly reduce the network distance between known COPD GWAS gene products; (2) 5.2% of the newly identified interactors are differentially abundant between COPD cases and controls in lung proteomic data from the Lung Tissue Research Consortium (LTRC); and (3) 41 (14.3%) of the newly identified interactors were directly connected to more than one of the six COPD GWAS genes. We identified a tightly connected core network of COPD GWAS gene products and protein biomarkers with intersecting signals; this network consists of 8 proteins, including 5 of the 6 GWAS gene products (AGER, FAM13A, HHIP, FBXO38, and MFAP2) as well as CAVIN1, TGM2, and HSPA1A. Conclusions: By combining AP-MS experimental data, multiple types of "omics" data, and network analysis, we have constructed a disease network module for COPD. This module can be a foundation for understanding the collective influence of COPD GWAS gene products in disease pathogenesis.

systems biology↗

bioRxiv2026

Genome-resolved metagenomics reveals metabolically versatile Actinobacteria with extensive biosynthetic capacity in Hawaiian steam vents

Geothermally active lava caves and hydrothermal steam vents are chemically heterogeneous subsurface environments that harbor diverse microbial communities, yet the ecological and metabolic roles of Actinobacteria in these systems remain poorly characterized. Here, we reconstructed and analyzed 58 actinobacterial metagenome-assembled genomes (MAGs) representing five classes. Comparative genomic analyses revealed broad metabolic versatility, including pathways for amino acid biosynthesis and utilization, central carbon metabolism, fatty acid degradation, and potential chemolithotrophic processes involving carbon monoxide and sulfur compounds. The MAGs also encoded diverse carbohydrate-active enzymes and peptidases suggesting substantial capacity for complex organic carbon and protein utilization. Genome mining identified 219 biosynthetic gene clusters spanning terpenes, ribosomally synthesized and post-translationally modified peptides, nonribosomal peptide synthetases, polyketide synthases, and {beta}-lactones, highlighting extensive secondary-metabolite biosynthetic potential. Antibiotic resistance-associated genes were also detected in several MAGs, with glycopeptide-resistance-associated van genes particularly prevalent among Thermoleophilia. Sequence similarity network analysis revealed that several van-associated proteins from Thermoleophilia and UBA4738 shared substantial sequence similarity with homologs from other environmental Actinobacteria, suggesting broad conservation of these protein families. Collectively, these findings reveal metabolically and functionally diverse Actinobacteria with the genomic potential to participate in carbon and nutrient cycling, microbial interactions, secondary metabolism, and antibiotic resistance in geothermally active ecosystems.

microbiology↗

bioRxiv2026

A Tumor-Pituitary-Stromal Axis Promotes Colorectal Cancer Progression

Peripheral tumors can reprogram systemic physiology, yet how they engage central endocrine organs to remodel the local tumor microenvironment is unknown. We found that colorectal cancer (CRC) elevates circulating BMP2, which activates pituitary signaling and enhances endocrine output. Increased pituitary-derived growth hormone acted on growth hormone receptor-expressing cancer-associated fibroblasts (CAFs), inducing a p21-dependent senescence-like state and tumor-promoting secretory program. Disruption of pituitary BMP signaling or fibroblast GHR reduced tumor growth, whereas fibroblast-specific depletion of p21 abrogated GHR-dependent tumor promotion. In patients with CRC, elevated pituitary metabolic activity correlated with stromal p21 expression and tumor burden. Our study reveals a tumor-pituitary-stromal signaling axis through which a peripheral malignancy recruits a central endocrine organ to remotely shape its own microenvironment and promote disease progression.

cancer biology↗

bioRxiv2026

ERBB2 establishes the timing of progesterone priming required for uterine receptivity

Progesterone is essential for establishing uterine receptivity and sustaining early pregnancy. However, how the uterus acquires the competence to respond to progesterone at the appropriate time remains poorly understood. Here we show that ERBB2 governs the timing of progesterone responsiveness required for uterine receptivity. In endometrial samples from women with recurrent spontaneous abortion (RSA), ERBB2 is downregulated specifically in the stromal compartment, accompanied by a parallel reduction in stromal progesterone receptor (PGR). Uterine deletion of Erbb2 in mice blunts the rise of stromal PGR at the onset of the receptive window, deferring implantation and impairing decidualization despite normal circulating progesterone and estrogen. Epithelial deletion has no such effect, placing the requirement in the stroma. High-dose progesterone fails to rescue the phenotype; advancing progesterone priming by 24 hours restores stromal PGR, HAND2, Bmp2, and on-time implantation. Collectively, the present study provides evidence that ERBB2 directs the temporal window of stromal progesterone priming by setting the threshold for PGR induction, thereby synchronizing the maternal-fetal interface and safeguarding pregnancy.

developmental biology↗

bioRxiv2026

Demography and selection shape gene loss in freshwater sticklebacks - A pangenomic perspective

The extent of within-species variation in genome assembly size, and the relative contributions of neutral vs. adaptive processes to it, remains poorly understood. We assembled the pangenome for the nine-spined stickleback (Pungitius pungitius) from whole-genome resequencing data obtained from range-wide sampling of 1,478 individuals across 47 populations. Our analysis recovered approximately 234.15 Mb of novel sequence and identified 11,151 protein-coding genes annotated in non-reference-represented regions, which accounted for 28.8% of all annotated pangenomic genes, corresponding to a 40.6% expansion relative to the 27,450 protein-coding genes annotated in the updated v8 reference genome. The identified presence/absence variation (PAV) genes exhibited a higher incidence of deleterious and non-synonymous mutations than core genes. Freshwater populations with smaller effective population sizes had lost more genes and had smaller assembly sizes relative to larger marine populations. Larger genome sizes of marine populations were primarily associated with higher transposable element content. Across gene categories, patterns of molecular variation were consistent with weaker purifying selection in freshwater populations than in marine populations, while recurrent freshwater-biased loss of a subset of PAV genes suggested that gene loss is not entirely random. Together, these results indicate that demographic history explains the overall pattern of gene loss, whereas negative natural selection may contribute to repeated loss of a subset of genes.

genomics↗

bioRxiv2026

Chemical Descriptors and Deep Learning Embeddings for Scoring de novo Peptide Designs

Peptides occupy a valuable niche between small molecules and biologics, but the clinical translation of de novo peptide designs requires rigorous scoring to simultaneously optimise target binding affinity alongside multiple developability traits, including stability, membrane permeability, aggregation propensity, and non-fouling behaviour. Here, we evaluate two distinct approaches for scoring these candidates: classical chemical descriptors and modern deep learning representations derived from protein language and folding models. Assembling nine public datasets spanning five developability traits and four binding-affinity endpoints, we find sequence-derived chemical descriptors alone contain sufficient information to predict developability task labels effectively. Given their drastically lower computational cost and higher interpretability, classical machine learning models trained on these simple descriptors frequently match or approach the performance of complex deep learning architectures, emerging as a highly efficient and interpretable alternative for high-throughput scoring. Finally, for scoring binding affinity, we demonstrate that Boltz-2 pair representations capture the most information among the tested representations; however, the model's predictive power is confounded by a significant bias from the molecular weight of the peptides. Together, these results establish a comprehensive assessment of state-of-the-art methods for predicting both peptide developability and binding affinity, highlighting the enduring value of interpretable chemical descriptors alongside deep learning in the scoring and selection of de novo peptide designs.

bioinformatics↗

bioRxiv2026

Large phospholipid-number asymmetry is not required to reproduce plasma membrane physical properties

Plasma membranes are compositionally asymmetric, but whether this lipid-type asymmetry is accompanied by a substantial phospholipid-number imbalance between leaflets remains debated. Here, we use microsecond all-atom molecular dynamics simulations to compare human red blood cell plasma membrane models with either strong phospholipid-number asymmetry and exoplasmic cholesterol enrichment or near-symmetric phospholipid and cholesterol numbers but preserved lipid-type asymmetry. Both asymmetric models reproduce a densely packed, ordered exoplasmic leaflet and a more fluid cytoplasmic leaflet. Strong phospholipid-number asymmetry, however, drives extensive cholesterol enrichment in the exoplasmic leaflet and amplifies membrane asymmetry, leading to large cholesterol-rich clusters, enhanced shallow hydrophobic exposure, reduced exoplasmic lipid mobility, lower ethanol permeability, and increased area compressibility. Comparison with available diffusion and alcohol-permeability measurements indicates that the strongly asymmetric model overestimates the immobilization and barrier properties of the exoplasmic leaflet, whereas the near-symmetric-number model better captures these dynamic observables. Our results suggest that lipid-type asymmetry is sufficient to reproduce many physical hallmarks of plasma membranes without requiring a large phospholipid-number imbalance.

biophysics↗

bioRxiv2026

Deep-layer cortical tracking abruptly collapses in the absence of language comprehension

The brain builds meaning from speech in stages, transforming acoustic input into linguistic comprehension. Yet where comprehension separates from general acoustic processing has been difficult to localize, because the two are tightly entangled in continuous speech. Here we align the activity of ~145,000 individual artificial neurons of an audio large language model with high-resolution magnetoencephalography, using single-neuron interpretability to track, layer by layer, which computations the cortex follows during natural listening. Contrasting native listeners with listeners hearing an unfamiliar language, while holding the acoustics physically identical, we find that comprehension selectively sustains brain-model alignment through the model's deep layers. Without comprehension, alignment in the acoustic encoder is preserved but deep-layer tracking abruptly collapses, and the sparse alignments that survive correspond only to physical acoustic features. This pinpoints the stage at which comprehension departs from perception and yields an interpretable, non-invasive index of whether speech is understood.

neuroscience↗

bioRxiv2026

Targeting fibroblast activation suppresses chemotherapy-induced pro-invasive remodeling of the extracellular matrix

Triple-negative breast cancer (TNBC) is an aggressive subtype of breast cancer that accounts for nearly 20% of breast cancer diagnoses. Despite recent approvals of novel therapeutic agents for TNBC, neoadjuvant chemotherapy remains part of standard of care treatment for TNBC patients. However, approximately 60% of TNBC patients who received neoadjuvant chemotherapy fail to achieve pathological complete response and face significantly elevated risks of recurrence and distant metastasis. We have shown that chemotherapy can induce significant changes in TNBC tumor extracellular matrix (ECM) that drive tumor cell invasion and may contribute to recurrence. Here, we investigate how chemotherapy drugs can impact resident fibroblasts, the major generators of ECM in the breast tissue. We find that chemotherapy activates fibroblasts and that doxorubicin specifically induces the production of pro-invasive ECM by fibroblasts via TGF-{beta}/Smad3 signaling. An FDA-approved tyrosine kinase inhibitor nintedanib suppressed doxorubicin-induced fibroblast activation and pro-invasive ECM production by fibroblasts in vitro. Additionally, in a TNBC mouse model, nintedanib reduced fibroblast activation markers in tumors, suppressed tumor cell motility in the tumor ECM, and reduced lung metastasis. These findings demonstrate a novel mechanism by which chemotherapy can induce fibroblast activation and imply that targeting fibroblast activation reduces chemotherapy-induced metastasis in TNBC.

cancer biology↗

bioRxiv2026

Transthyretin Contributes to Hippocampal Blood-Brain Barrier Recovery Following Intestinal Inflammation, with Reduced Expression in Alzheimer's Disease

Background: The blood-brain barrier (BBB) is increasingly recognized as an active immunoregulatory interface that responds dynamically to systemic inflammation. As intestinal inflammation can influence brain homeostasis through the gut-brain axis, the endogenous mechanisms that preserve BBB integrity during gut-derived inflammatory stress remain poorly understood. Methods and Results: We combined dextran sulfate sodium-induced colitis, fecal microbiota transplantation, aged mice, APP/PS1 mice, human Alzheimer's disease (AD) brain tissue, single-cell RNA sequencing, in vivo BBB permeability assays, and gain- and loss-of-function approaches to investigate adaptive neurovascular responses to gut inflammation and their mechanism in maintaining BBB integrity. Intestinal inflammation induced a region-specific adaptive response characterized by increased hippocampal vascular remodeling and partial restoration of BBB integrity following the initial inflammatory insult. Single-cell transcriptomic analysis identified a transthyretin (TTR)-enriched vascular-associated microglial state accompanying these neurovascular changes. Functional studies demonstrated that TTR contributes to maintaining BBB integrity by promoting endothelial homeostasis and limiting endothelial endocytosis, consistent with modulation of FcRn-associated transport pathways. This adaptive neurovascular response was progressively attenuated in aged mice and APP/PS1 mice and was accompanied by reduced vascular TTR expression in human AD brains. Conclusion: These findings identify TTR as a contributor of adaptive neurovascular homeostasis during gut-derived neuroinflammation. Impairment of this homeostatic response with aging and AD may contribute to persistent BBB dysfunction and chronic neuroinflammation, highlighting neurovascular resilience as a potential therapeutic target.

neuroscience↗

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

TomatoPGFM: A graph-conditioned foundation model for tomato pangenomes

Most genomic foundation models are pretrained on independent linear assemblies and therefore do not explicitly represent population-level segment sharing or local graph connectivity. We developed TomatoPGFM, a graph-conditioned model pretrained on 54.65 Gb of sequence from 66 tomato (Solanum spp.) accessions. Sequence tokens were conditioned on pangenome node attributes and local adjacency, and the model was optimised using masked language modelling and graph-feature reconstruction. To evaluate model responses to graph-conditioned input, we compared aligned, shuffled and disabled graph inputs in 25,000 windows from the training panel. Sequence-aligned graph input produced lower masked language modelling loss than graph-off at all five curriculum stages in both training-panel strata, while the shuffled perturbation generally yielded intermediate losses. We then assessed sequence-only transfer in Solanum sitiens LA1974 and S. lycopersicum MicroTom, neither of which was used for graph construction or pretraining. Frozen-probe AUROC values for gene-versus-intergenic and coding-sequence-versus-intergenic classification ranged from 0.8489 to 0.9593. TomatoPGFM produced higher AUROC point estimates than DNABERT-2 in all four comparisons. Enabling the zero-feature GraphAdapter pathway with adjacency messaging disabled changed throughput by less than 1% at 512-2,048 positions under the tested configuration. Together, these results show that TomatoPGFM responds consistently to sequence-aligned pangenome context in training-panel sequences and provides informative sequence representations for genic-region classification in accessions excluded from graph construction and pretraining.

bioinformatics↗

bioRxiv2026

Structural basis for catalytic and inhibitory divergence between archaeal and bacterial ammonia monooxygenases

Ammonia oxidation initiates nitrification and is closely linked to microbial N2O production. Ammonia monooxygenase (AMO) catalyzes the first and rate-limiting step of nitrification and is widespread across evolutionarily distinct ammonia-oxidizing archaea (AOA) and bacteria (AOB). The ocean is the largest biome for AOA and AOB, which have distinct ecological niches and markedly different sensitivities to nitrification inhibitors. However, the lack of archaeal AMO structures and inhibitor-bound AMO complexes has hindered mechanistic understanding of the architectural, catalytic, and inhibitory divergence between these two enzyme systems. Here, we report high-resolution cryo-electron microscopy (cryo-EM) structures of marine archaeal AMO captured in active and inactivated states within its native membrane environment, together with inhibitor-bound structures of estuarine bacterial AMO. Archaeal AMO forms an unexpected cup-shaped homotrimer composed of eight subunits per protomer and exhibits substantial architectural divergence from bacterial AMO. Integrated structural, biochemical, kinetic, and computational analyses reveal distinct periplasmic architectures, copper-center organization, and hydrophobic channels between archaeal and bacterial AMOs for ammonium acquisition, catalysis and inhibitor response. These findings provide a structural and mechanistic framework for understanding how archaeal and bacterial AMOs have diverged to distinct ammonia-oxidizing strategies and inhibitor susceptibilities across environmentally important ammonia oxidizers.

molecular biology↗

bioRxiv2026

OsRAD23a negatively regulates salt tolerance and phosphorus uptake in rice

Salinity stress affects rice productivity due to reduced growth and sodium ion toxicity. Previously, we identified a splice variant of RADIATION SENSITIVE23a (RAD23a) as the potential basis for variation in salt-tolerance in rice germplasm. RAD23 is a known moonlighting protein associated with protein degradation. To validate the role of RAD23a in salt stress response, we characterized gene edited mutant lines that targeted the UBL and UBA2 domains of this protein. Mutation in either domain promoted shoot growth under saline and control conditions. The mutants also differed from wildtype plants in Na and K accumulation in roots and shoots under salt stress. Transcriptome analysis of mutants versus wildtype showed differential transcript abundance of multiple inorganic phosphate (Pi) starvation related genes, including OsSPX2 and OsPHO2. As a result, mutants accumulate higher Pi compared to wildtype plants. The two allelic groups for RAD23a locus also differ in root and shoot phosphorus (P) content. Further, we show that RAD23a interacts with OsSPX2, a negative post-translational regulator of OsPHR2, the master regulator of Pi starvation response. Mutants have higher shoot growth and Pi levels under low Pi conditions, linking enhanced growth of mutants to increased Pi uptake. The UBA2 domain specific mutants have higher single grain weight and per plant grain weight than wildtype. In summary, we show that the RAD23a regulates differential growth, salt response and Pi uptake in rice in a domain-specific manner supporting the moonlighting roles of RAD23a in salt tolerance and phosphorus-dependent shoot growth.

plant biology↗

bioRxiv2026

MultiFlow: coupled flow matching for predicting single-cell multiomic perturbation responses in unseen cellular contexts

Predicting cellular responses to perturbation requires resolving coordinated changes across molecular layers, yet most single-cell perturbation models focus on transcriptional responses alone. Here we present MultiFlow, a coupled flow-matching framework that unifies generation and perturbation prediction of paired gene expression and chromatin accessibility. By learning coupled RNA-ATAC flows conditioned on perturbation and control-derived cellular-state representation, MultiFlow enables prediction of coordinated multiomic responses in unseen cellular contexts. Across multiomic generation benchmarks, MultiFlow accurately reproduced paired RNA-ATAC states and their population distributions. In multiomic perturbation benchmarks, MultiFlow achieved the strongest overall performance in predicting both gene-expression and chromatin-accessibility responses, outperforming competing modality-specific perturbation-prediction methods. Joint multiomic modeling further preserved perturbation-induced RNA-ATAC coordination, including concordant peak-gene effects and cross-modal cellular neighborhood structure. These results establish coupled flow matching as a unified generative framework for modeling paired multiomic states and predicting coordinated perturbation responses across cellular contexts. Code and tutorial for MultiFlow are available at https://github.com/liuq-lab/MultiFlow.

bioinformatics↗