Biology subjects

Zhang, Z.

Publications and source records attributed to Zhang, Z..

At least 19 recordsLinked to original sources

bioRxiv2026

Histone modifications define distinct nucleosome positions in cell-free DNA

Circulating cell-free DNA (cfDNA) fragmentation patterns reflect nucleosome positioning. We asked whether variation in nucleosome positions in cfDNA can reveal chromatin states and associated histone modifications. Using cell-free chromatin immunoprecipitation sequencing (cfChIP-seq) in plasma from patients with prostate cancer, we identified 35,120 modification-associated positions (MAPs), loci where nucleosomes bearing active (H3K4me2) and repressive (H3K9me3) histone marks occupy reproducibly offset positions. These offsets recurred across histone marks, an independent cfChIP-seq cohort, and nuclease-digested chromatin from cell lines. MAPs were enriched in transcribed genes but depleted near transcription start sites and CTCF-binding sites where nucleosome positioning is constrained. Across 2,535 cfDNA sequencing datasets, nucleosome protection at a subset of MAPs correlated with tumor fraction (TF), reflecting differences in histone modification status between cancer-derived and non-malignant chromatin. In targeted sequencing panels, a 50-MAP model using fragments from < 0.001% of the genome distinguished samples with tumor fractions [≥]3% vs 0% (AUCs = 0.91-0.96 across prostate, breast, and lung cancers) and outperformed a state-of-the-art fragmentomics-based tumor fraction predictor. MAP-derived estimates correlated with reference TFs (r = 0.89) and longitudinal changes in TF (r = 0.91). MAP protection at ASCL1 binding sites identified neuroendocrine transformation of prostate cancer from cfDNA. These findings establish histone-modification-associated nucleosome positioning as a mutation-independent readout of tumor fraction and cancer state from routine cfDNA sequencing.

cancer biology↗

bioRxiv2026

Building virtual cell dynamics that evolve with biological measurements

Virtual cell models built from single-cell measurements need to predict how cells change over time, but new biological measurements often reveal dynamical relationships that existing models cannot express. Translating such measurements into dynamics therefore requires deciding what biological relation should be modeled, where it should enter the formulation and how its recovery should be validated. Here we present CellCompass, an agentic AI system that couples a scientific agent with an extensible dynamical engine for modeling state transitions and population change, along with fate and molecular readout. CellCompass reuses existing formulations when sufficient and designs, implements and refines new ones when experimental measurements introduce additional constraints. To assess this capability, we developed DynBench100, a benchmark of 100 simulated single-cell time-course scenarios with hidden dynamical ground truth. CellCompass achieved the highest aggregate performance among the tested general-purpose and biomedical agents across six dynamical tasks. Across four real single-cell studies, it retained an existing velocity-growth model to reconstruct context-dependent epithelial-mesenchymal transition and developed three new dynamical formulations for distinct experimental measurements. In drug-response data, incorporating metabolic RNA labeling reduced pulse-control identifiability error by 31.1%, revealing staged response programs and generating testable predictions for perturbing a candidate adaptive program. In hematopoietic lineage tracing, incorporating clonal barcodes improved clone-fate agreement by 54% over the strongest dynamical baseline, identifying early lineage priming among transcriptionally similar progenitors. Finally, in time-resolved CITE-seq data, CellCompass reduced future surface-protein distribution error by 13.2% for RNA-initialized trajectories, uncovering lineage-specific temporal offsets between transcriptional and protein programs. Together, CellCompass provides a framework in which virtual-cell dynamics can expand with the biological measurements used to constrain them.

bioinformatics↗

bioRxiv2026

Establishment of a cochlear lymphatic vascular network and immune surveillance system following cochlear implantation

Hearing loss is a highly prevalent sensory abnormality often addressed with cochlear implants (CI) that provide auditory rehabilitation for deaf patients. However, a universal inflammatory foreign body response (FBR) to the implanted materials negatively impacts CI outcomes. The mechanistic basis for immune cell infiltration and propagation of the FBR directed against CIs is not defined. Using mouse CI models and quantitative, highly tractable reporters for lymphatics, blood vessels, and immune cells, we identify inner ear lymphatic vessels and changes to the immune surveillance system following cochlear implantation. Histopathological analysis of implanted temporal bones from human subjects shows that both blood vessels and lymphatics are evident within the scala tympani of the implanted cochlea. Our data reveal a rich lymphatic vascular network in the inner ear, closely associated with the perilymphatic and endolymphatic spaces. Lymphatics proliferate within the perilymphatic space of the implanted cochleae and appear to traffic immune cells. Together, our data provide direct evidence that concurrent proliferation of blood vessels and recruitment of blood-borne immune cells comprise an 'immune surveillance system' in the implanted cochleae that promotes the FBR. Our findings also highlight potential opportunities to limit adverse FBR outcomes following cochlear implantation.

systems biology↗

bioRxiv2026

ScienceRSI: Recursive Scientific Self-Improvement

Scientific AI can improve cumulatively when capabilities developed in one model become useful learning signals for the next. We introduce ScienceRSI, a framework for recursive scientific self-improvement organized around specialize, verify, transfer, and reuse. Compact experts develop complementary scientific capabilities; shared evaluation identifies their useful contributions; a target model integrates those contributions and can serve as the starting point for further specialization. This organization connects reasoning, model scaling, and iteration through reusable scientific capabilities. We evaluate its transfer step in 4B[->]4B, 4B[->]8B, and 8B[->]27B configurations. ScienceRSI ranks first among the compared methods on all five science benchmarks for the 4B and 8B students. Relative to rubric-based reinforcement learning, scientific-reasoning averages improve by 2.0 and 1.9 percentage points, and domain-science averages by 3.3 and 2.1 points, respectively. In a separate cross-generation 27B extension, reported scores improve over Vanilla on all seven benchmarks, with the unweighted mean rising from 87.3% to 94.0%. These results support verified capability transfer across model scales as a practical foundation for ScienceRSI. Accumulating these gains over repeated rounds is the next step toward sustained recursive scientific improvement.

bioinformatics↗

bioRxiv2026

Stromal control of sympathetic rewiring in metastatic lymph nodes

Metastatic lymph nodes (LNs) are dynamically remodeled tissues, but this process has been understood mainly in immune and stromal terms. We show that tumor-driven LN remodeling also involves adaptive neural rewiring. In a mouse model of nodal metastasis, sympathetic fibers, but not sensory fibers, progressively expanded during LN remodeling and tracked with metastatic progression. Fibroblastic reticular cells (FRCs) adopted a neuro-supportive program characterized by increased hepatocyte growth factor (HGF) production, while local stromal HGF acted through MET on sympathetic neurons to stimulate neurite outgrowth. Increased substrate stiffness enhanced FRC activation and HGF production, suggesting that mechanical remodeling may reinforce the neuro-supportive stromal state. Inhibition of HGF signaling or ablation of sympathetic innervation reduced LN enlargement and nodal tumor burden. These findings identify neural plasticity as a functional component of the metastatic LN niche and reveal a stromal/mechanical/neural framework that supports metastatic LN progression.

cancer biology↗

bioRxiv2026

RDEmbed: A Sentence-Embedding Framework for Rare Disease Phenotype Representation and Phenotypic Similarity Analysis

Accurate phenotypic characterization of patients with rare diseases is essential for improving diagnosis, patient matchmaking, and gene discovery. Conventional ontology driven approaches based on the Human Phenotype Ontology (HPO) enable structured phenotype comparison but often limited by the incomplete annotations and reduced ability to capture the contextual information contained in free-text clinical narratives. We present RDEmbed, a sentence embedding framework that learns patient phenotype representations directly from rare disease case reports for similarity analysis. RDEmbed encodes narrative case reports into high-dimensional vectors to quantify phenotypic similarity. Unlike ontology-based approaches that rely on manually curated HPO annotations, RDEmbed encodes narrative case reports into dense semantic embeddings that capture complex phenotypic relationships from unstructured text. We further introduced retrieval-augmented embedding refinement (RAER), which enhances embedding representations by incorporating external domain knowledge at inference time without task-specific fine-tuning. RDEmbed was evaluated against HPO-based similarity methods (OWLSim, simGIC) using simulated datasets, published rare disease case reports and a curated case study dataset comprising 156 patient abstracts presenting six rare disease groups. Across the real world datasets and curated cohort. RDEmbed consistently outperformed ontology based approaches, improving AUROC and AUPRC by more than 5%. Embedding-based representations also produced more distinct disease-specific clustering in t-SNE visualizations and identified twice as many true positive rare disease pairs as ontology based methods (8 versus 4). Incorporating RAER further improved precision while maintaining high sensitivity compared with zero-shot embeddings. These results demonstrate that sentence embeddings provide an effective alternative to ontology-based phenotype representations by learning directly from free-text clinical narratives. RDEmbed enables more accurate phenotypic similarity analysis while reducing dependence on manually curated HPO annotations and offers a scalable framework for rare disease diagnosis, patient matchmaking, and gene discovery. RDEmbed is publicly available at https://github.com/xhwang2017/RDEmbed.

bioinformatics↗

bioRxiv2026

Human brain white matter charts in international populations

Brain charts provide normative references for identifying neurodevelopmental milestones and clinically meaningful deviations. Existing white matter (WM) charts are derived predominantly from European and North American populations, limiting generalizability. Here, we established the first set of population-specific WM charts of 198 morphological and microstructural phenotypes for both Chinese (N=19,215) and Western (N=15,584) populations across the human lifespan. Developmental trajectories and spatial organization of WM maturation and degeneration were broadly conserved across the populations, whereas maturational timing varied across pathways and metrics. The Chinese population showed earlier maturation in several projection and sensorimotor pathways, but more protracted maturation in higher-order association and limbic pathways. Chinese population-matched charts more accurately predicted WM phenotypes and detected more extreme disease-related deviations in brain health in the Chinese population than Western-derived models. Individual deviation patterns further discriminated patients from healthy controls and predicted cognitive performance and plasma biomarkers. Our findings reveal population-specific WM maturation and show that reference-population mismatch can reduce the accuracy and clinical sensitivity of normative brain charts.

neuroscience↗

bioRxiv2026

Distance-AF Multimer: A Multi-Step Approach for Protein Complex Structure Modeling with User-Defined Distance Constraints

The three-dimensional structure of proteins provides essential insight into the molecular mechanisms underlying their function. Many biological activities are mediated through protein interactions, making accurate structural determination of protein complexes important. Despite significant advances in computational protein structure prediction, there are still many cases where protein complexes are not correctly assembled. In particular, there are cases where experimental data, such as distances between residue pairs in interacting proteins are available, yet existing prediction methods fail to produce models consistent with these data. Here, we present Distance-AF Multimer, a method that guides and refines AlphaFold-Multimer docking predictions using user-provided residue pair distance information. This approach extends our previously successful data-guided single-protein modeling method, Distance-AF. Distance-AF Multimer operates in a two-step process, starting from multiple initial conformations and iteratively refining them to satisfy the provided distance constraints. We evaluated Distance-AF Multimer on protein-complex benchmarks spanning different chain counts and docking errors, demonstrating competitive or superior performance compared to other existing deep-learning-based structure prediction methods. In addition to dimer cases, we also demonstrated the potential of Distance-AF Multimer for large-scale protein assemblies containing multiple chains, including those with more than 5 chains.

bioinformatics↗

bioRxiv2026

MOFI reveals dynamic multi-omics coordination during cell fate progression

Cell fate decisions are shaped by dynamic coordination across transcriptomic, chromatin, and protein layers. Time-series single-cell multi-omics assays now enable these molecular layers to be profiled across developmental stages, yet each cell is still observed only once, and the continuous multi-omics trajectories connecting discrete snapshots remain unresolved. A central challenge is therefore to jointly integrate paired omics measurements while reconstructing their shared developmental progression. Here we present Multi-Omics Fate Inference (MOFI), a dynamically constrained multi-omics integration framework designed to reconstruct continuous cell fate transitions from paired multi-omics snapshots sampled across multiple time points. MOFI jointly models omics-specific state dynamics, cross-omics correspondence, and shared population growth within a unified framework, enabling cross-omics perturbation response analysis and identification of candidate cross-omics transformation programs. Across simulated benchmarks and paired RNA-ATAC/RNA-protein time-series datasets, MOFI reconstructed continuous fate transitions, population dynamics and cross-omics changes consistent with observed measurements. In kidney organoids, MOFI captured coordinated RNA-ATAC progression of nephron lineage specification. In hematopoietic progenitors, MOFI predicted perturbation-responsive lineage redistribution and nominated multi-omics transformation programs linking transcriptional modules with protein-level responses. In neural developmental systems, MOFI resolved gradual brain organoid maturation and interneuron lineage allocation during human cortical development. Together, MOFI provides a dynamics-grounded method for converting discrete paired multi-omics snapshots into interpretable models of cell fate progression, perturbation response, and cross-omics transformation analysis.

bioinformatics↗

bioRxiv2026

Self-extending DNA recording with stochastic and programmed writing

Recording cellular histories requires new information to accumulate without erasing earlier records. Here we introduce Infinite Printer, a CRISPR recording architecture that couples self-extension to regeneration of a writable terminal context. A self-targeting prime-editing guide RNA (stpegRNA) directs repeat extension, while base editing generates diversity in situ without a separate insertion template for each character state. Earlier sequence remains upstream of the renewed terminus, linking substrate growth with record retention. Auxiliary pegRNAs specify payloads within a write-renew-write architecture. After three rounds of payload input, 4,436 distinct complete CCA-based records contained identities from all three pools. Longer AC/AT records retained shared upstream marks, and full-length sequence analysis resolved additional endpoint distinctions. Joint single-cell readout recovered 1,549 distinct records and 7,669 combinations across 12,302 transcriptome-linked cell labels from a single stpegRNA design. Spatial-transcriptomic readout recovered spacer records at 11,330 capture spots, linking record content to tissue coordinates and gene-expression profiles. Infinite Printer unites substrate renewal, sequence diversification and programmed writing with cell- and tissue-linked readout, providing a foundation for future lineage recording across large cell populations during mammalian development.

synthetic biology↗

bioRxiv2026

Protein Scaffolds Modulate the Electronic Structure of a Common Ketyl Radical Intermediate in Thiamine Pyrophos-phate-Dependent Radical Biocatalysis: Insights from EPR and ENDOR Spectroscopy

New-to-nature radical biocatalysis has emerged as a powerful strategy for asymmetric C--C bond formation, yet the enzyme-bound radi-cal intermediates that underpin these transformations have remained largely uncharacterized. Here, we combine photo biocatalytic reaction development with multifrequency electron paramagnetic resonance (EPR), electron-nuclear double resonance (ENDOR) spectroscopy, and density functional theory (DFT) calculations to elucidate the ketyl radical intermediates formed during thiamine pyrophosphate (TPP)-dependent radical biocatalysis. We first develop an oxidative photobiocatalytic cross-coupling of aldehydes with alkyltri-fluoroborates catalyzed by engineered benzoylformate decarboxylase (BFD) variants, providing enantioenriched -chiral ketones through a new oxidative radical manifold inaccessible to benzaldehyde lyases (BALs). Freeze-quench multifrequency EPR and ENDOR spectroscopy directly identify the catalytically generated TPP-bound ketyl radical and reveal that this common radical intermediate adopts distinct electronic structures in BFD and BAL despite their identical chemical composition. Analysis of 1H, 19F, 31P, and 14N hyper-fine interactions, together with DFT calculations, demonstrates that enzyme-specific active-site conformations redistribute spin density between the TPP cofactor and substrate fragment. Greater substrate-centered spin delocalization in BFD provides a mechanistic basis for its superior radical C--C coupling activity relative to BAL. These results establish a direct relationship between protein-controlled electronic structure and catalytic function in TPP-dependent radical biocatalysis and demonstrate that engineering the electronic properties of a common radical intermediate represents a useful strategy for controlling enzyme reactivity in radical biocatalysis.

biochemistry↗

bioRxiv2026

Cortex-wide single-cell activity mapping in freely moving mice

How the cortex supports complex, naturalistic behavior can be revealed only in freely moving animals. Although head-mounted microscopes promise access to the neural dynamics underlying such behavior, fundamental trade-offs among field of view, spatial resolution, device weight, and background contamination have prevented cortex-wide single-neuron recording, leaving distributed cortical coordination largely inaccessible. To meet these challenges, we introduce COMET (Cortex-wide Observational Miniature Epifluorescence Technique), a 3.75-g head-mounted mesoscope that integrates an aberration-controlled wide-field optical design, uniform high-efficiency excitation and single-coaxial power-and-data transmission with an end-to-end denoising and signal-extraction pipeline. This enables cortex-wide calcium imaging at single-cell resolution in freely moving mice. COMET captures activity from >15,000 layer 2/3 neurons per 5-min recording over fields of view up to 10 mm in diameter while preserving natural behavior. Benchmarking against benchtop two-photon microscopy demonstrates high spatial concordance of detected neurons and matched event-level calcium dynamics. We apply COMET to social interaction to quantify region-specific engagement and to an auditory trace fear-conditioning task that requires natural movement and behavioral expression, revealing time-ordered changes in regional firing-rate structure and functional connectivity preceding a global conditioned response. COMET provides an open, lightweight hardware-software framework for linking single-neuron activity to cortex-scale network dynamics during freely expressed behaviors under changing internal or external conditions.

neuroscience↗

bioRxiv2026

Non-canonical IL-2 signaling facilitates peripheral regulatory T cell differentiation

Non-canonical IL-2 signaling, termed Monoclonal antibody or Matrix-directed Alternative Cytokine Signaling (MACS), is mediated by a CD25-chemokine receptor complex and operates in concert with the canonical JAK-STAT5 pathway; however, its role in T cell homeostasis remains poorly defined. Here, we uncover a role for MACS in regulatory T cell (Treg) differentiation and functional specification. We generated Il2E66K/E66K mice in which CD25-CCR7 assembly is disrupted and IL-2-induced STAT5 phosphorylation is preserved. Immunophenotyping analysis by flow cytometry revealed that MACS deficiency reduces the accumulation of Nrp1- Foxp3+ Tregs in gut-associated lymphoid tissues from 24-weeks-old Il2E66K/E66K mutant mice. Single-cell transcriptomics, functional suppression assays, and in vitro differentiation revealed that MACS engagement drives Treg differentiation and sustains effector Treg programs - marked by robust IL-10 expression - in part via an AP-1-dependent transcriptional network. These findings establish IL-2 MACS as a distinct signal directing peripheral immune tolerance and highlight a potential node for targeted immunomodulation.

immunology↗

bioRxiv2026

Hidden grassland productivity from overlooked herbivore impacts

Grasslands cover approximately 40% of Earth's land surface and contribute substantially to terrestrial primary production. Despite widespread grazing, field estimates of grassland aboveground net primary productivity commonly overlook two grazing-related processes: biomass consumed by herbivores and post-grazing compensatory growth. The magnitude and direction of the resulting biases remain unclear. Here we quantified these biases using a 3-year yak grazing experiment on the Tibetan Plateau, a meta-analysis of 152 grazing experiments across China, and a global synthesis of 212 paired observations from 58 studies. Overlooking biomass consumed by herbivores consistently underestimated productivity by 34-51%. By contrast, overlooking grazing-induced compensation generated smaller but more variable biases, ranging from 29% underestimation to 19% overestimation, mainly reflecting contrasting compensatory responses under rotational versus continuous grazing. These findings show that overlooking grazing-mediated processes introduces substantial, process-dependent biases in grassland productivity estimates, underscoring the need to incorporate these processes into ecosystem and climate assessments.

ecology↗

bioRxiv2026

A telomere-to-telomere genome of Holstein cattle resolves chromosome-end architecture and the genetic basis of telomere length

Telomeres safeguard chromosome-end integrity, yet population-scale dissection of telomere-length genetics has been largely restricted to humans. Here, we report a 3.01-Gb telomere-to-telomere (T2T) genome of Holstein cattle that resolves repeat-rich chromosome ends and centromeres and enables telomere-length quantification at single-chromosome-end resolution. Holstein chromosome ends show marked telomere-length asymmetry and end-specific variation in telomere variant repeats, subtelomeric tandem repeats, GC/CpG composition, and DNA methylation. Using sequencing-depth-corrected telomere-length phenotypes from 9,591 cows, teloGWAS identifies major loci encompassing RTEL1, CTC1, and TERT, while multi-tissue eQTL, sQTL, TWAS, and co-expression analyses prioritize CTC1 and implicate DNA replication, DNA repair, and stress-response networks. Together, these findings extend population-scale telomere genetics beyond humans and establish a sequence-resolved framework for chromosome-end biology and telomere-informed livestock genomics.

genomics↗

bioRxiv2026

Towards Robust Automated Mouse Scratching Behavior Detection for Itch Research

Automated detection of mouse scratching behavior could reduce manual annotation and is essential for preclinical itch research. However, robust performance across laboratories remains a challenge for current methods. We present Scratch-AID+, a framework dedicated to automated mouse scratching detection, including a diverse and progressive set of approaches: classic ResNet-based detectors, task-adaptive fine-tuning of video foundation models, and in-context learning with advanced multimodal large language models. To simulate real-world use cases, we benchmark on a public dataset and utilize additional cross-laboratory video recordings for independent testing. We find that classic baselines suffered severe performance degradation on external data under the realistic setting with sparse scratching events. In contrast, fine-tuned video foundation models successfully bridged this generalizability gap, achieving significantly better detection performance. Furthermore, frozen multimodal large language models demonstrated promising out-of-the-box ability using only few-shot demonstrations. We also examined the effects of different video preprocessing strategies. Our findings highlight the improved performance and generalizability of the adapted foundation models for automated scratching behavior analysis and shed light on the real-world deployment of such detectors. Our code is publicly available at https://anonymous.4open.science/r/scratch aid-plus/.

bioinformatics↗

bioRxiv2026

Filling the Missing Puzzle: Integrating Interaction and Evolutionary Information to Enhance Protein Fitness Prediction in Inverse Folding

Protein inverse folding, which infers compatible amino acid sequences from a given backbone structure, is central to rational protein design. Current methods rely largely on sequence recovery rate as the evaluation metric, emphasizing structural compatibility but seldom addressing function. Yet functional activity depends on the proper network of non-covalent interactions-protein-ligand contacts being a prime example. We argue that function-oriented protein design is governed by three complementary constraints: structural, interaction, and evolutionary, none of which are handled in a unified manner by existing pretrained models. To bridge this gap, we introduce zPolaron, a graph neural network built on protein-ligand complexes for functional protein design. zPolaron represents residues and ligand atoms as a heterogeneous graph and jointly encodes structural, interaction, and evolutionary information by fusing physical-interaction-enhanced graph representations with ESM-2 embeddings. Across multiple fitness prediction benchmarks, zPolaron outperforms all baselines, with gains evident across structural, interaction, and evolutionary dimensions. Wet-lab experiments further confirm its utility in functional enzyme screening and design: the optimized CalA6-D205A-S160G mutant achieves a 17.84-fold improvement in activity, while the CalB-I121M mutant yields a 1.91-fold increase. Together, these results show that integrating interaction and evolutionary information fills the missing puzzle of inverse folding, enhancing protein fitness prediction beyond structural compatibility. zPolaron is available at https://github.com/zelixirSH/zPolaron.

bioinformatics↗

bioRxiv2026

Chain of Structures: Connecting Reasoning and Generation for Protein Design

Scientific design requires reasoning in scientific spaces that represent the objects being designed. Building on Scinetics' Chain of Science Tokens, we present chain-of-structures (COS) as a concrete protein-design case study that connects reasoning and generation through explicit structural states. In this technical report, we apply COS to enzyme design, where catalytic function requires precise local geometry within a compatible global fold, making it a demanding test of reasoning from chemical requirements to protein structure. Starting from a reaction specification and target length, COS couples a language reasoning expert with an all-atom generation expert across four rounds: catalytic core, active site, global backbone, and complete sequence and structure. Each reasoning state directly conditions generation, and the resulting structure becomes context for the next decision. Discrete residue constraints and continuous latents connect the two experts, enabling structural supervision of the reasoning interface. On the evaluated benchmark, COS leads the compared baselines on five of six metrics, achieving 57.0% EC3 agreement and a TM-score of 0.729.

bioinformatics↗