Biology subjects

Xu, Z.

Publications and source records attributed to Xu, Z..

At least 37 records · Page 2Linked to original sources

bioRxiv2026

PathFold: Predicting the Entire Protein Folding Pathway from Protein Sequence Alone

Recent advances in protein structure prediction, exemplified by AlphaFold, have largely addressed the determination of static structures, one aspect of the protein folding problem. However, predicting folding pathways, by which proteins reach their native states, remains a significant challenge. Here, we present PathFold, a deep learning framework that predicts protein folding pathways directly from sequence information. PathFold leverages an AlphaFold-based module to extract structural information from the sequence and generates a progressive folding trajectory from an extended conformation using a diffusion model. By modeling the full trajectory, it enables prediction of folding intermediates and transition pathways, analogous to those observed in steered molecular dynamics (SMD) simulations. The predicted pathways reveal well-defined intermediates and sequential folding events, and show agreement with experimental folding data, including measured {Phi}-values.

bioinformatics↗

bioRxiv2026

PEG-Arginase 1: A Novel Therapy for Optic Nerve Injury

Traumatic optic neuropathy (TON) occurs due to direct or indirect injury to the optic nerve and is a significant cause of visual disability. So far, there is no effective treatment. The lack of understanding of the cellular mechanisms by which trauma induces inflammation and damage in retinal neurons is a critical knowledge gap in developing effective therapies. We have studied the role of the arginase 1 (A1) enzyme in this pathology. We have found previously that treatment with a long-acting form of human recombinant A1, pegylated A1 (PEG-A1) after optic nerve crush limits activation of retinal microglia and macrophages (M{Phi}) and reduces inflammation, thereby decreasing injury and protecting visual function. Here we report on studies designed to demonstrate the therapeutic efficacy of PEG-A1 in mouse models of direct and indirect TON and to elucidate the underlying mechanisms. We used ONC to model direct TON and sonication-induced trauma to the supraorbital rim to model indirect TON (SI-TON). At different times after injury, mice were treated with PEG-A1 which was delivered systemically by i.p. injection or locally by intravitreal injection. In order to assess the role of A1-induced activation of the ornithine/polyamine pathway in the protective effects of PEG-A1, some mice were treated with the ornithine decarboxylase (ODC) inhibitor, difluoromethylornithine (DFMO) immediately after the PEG-A1 treatment. Retinal function was determined by OptoMotry and electroretinography. Retinal injury and microglia/M{Phi} activation were assessed by immunofluorescence imaging. Expression of inflammatory cytokines was determined by Western blotting and quantitative RT PCR. Liquid chromatography mass spectrometry was used to analyze changes in arginase/ODC pathway metabolites. Results showed that PEG-A1 treatment improved neuronal survival and visual function whether delivered systemically or intravitreally. This neuroprotection was associated with decreased microglia/M{Phi} activation, decreased inflammatory cytokine expression, and increased formation of L-ornithine and putrescine. Furthermore, DFMO treatment blocked these effects, indicating that PEG-A1 limits retinal injury and preserves vision after ocular injury by activating ODC. ODC processes the arginase product L-ornithine to form polyamines which are known to promote reparative functions. Thus, PEG-A1 therapy offers a new strategy to limit trauma-induced vision loss and promote repair after TON.

cell biology↗

bioRxiv2026

Mechano-Piezoelectric F-Peptide Hydrogels Enable In Situ Stem Cell and Immune Niche Modulation for Cartilage Regeneration

Osteoarthritis, characterized by cartilage degradation and synovial inflammation, has spurred interest in mechano-piezoelectric bio-hydrogel therapies that can both relieve symptoms and reverse progression. However, current approaches lack sufficient piezoelectric output and dual cartilage/inflammation targeting. To address this, we demonstrated a mechano-piezoelectric peptide hydrogel composed of nanofibers integrating piezoelectric cues with mesenchymal stromal cells (MSCs) recruitment and PIEZO2 mechanosignaling. Molecularly, the hydrogel's seed peptide incorporated four functions: COL2A1 targeting, MMP-13 responsiveness, MSCs homing, and self-assembly. Overexpressed MMP-13 in the osteoarthritis niche triggers gelation, promoting MSCs recruitment and drug retention. Fluorination modulates hierarchical nanofiber assembly, enhancing mechanical and piezoelectric properties, as confirmed by morphological, biophysical, and computational analysis. The trifluoromethyl-modified, 4-octyl itaconate (4-OI) loaded formulation reverses osteoarthritis via PI3K/AKT activation and Wnt/{beta}-catenin suppression, as shown by improved Osteoarthritis Research Society International (OARSI) scores, bone microarchitecture, and cartilage matrix. This synergy of mechano-piezoelectric cues and 4-OI offers a clinically promising strategy for osteoarthritis.

biophysics↗

bioRxiv2026

Perturb-ME: Scalable mechanism discovery from phenotype-enriched genome-wide screens

Perturb-seq enables pooled genetic screens with rich single-cell profiling readouts, but genome-scale profiling remains costly and may not be associated with other established functional characteristics. Moreover, as screens grow in size and complexity, interpreting the resulting data comprehensively is challenging and slow. Here, we introduce Perturb-seq with Marker Enrichment (Perturb-ME), which combines genome-scale CRISPR screening, phenotype-based enrichment and multimodal single-cell profiling. Applied to MHC-I cell surface protein expression in melanoma, Perturb-ME profiled HLA-low and HLA-high cells with matched RNA, surface-protein and guide measurements. A regulatory model with 221 impactful regulators affecting 1,998 responsive genes recovered seven coherent co-functional regulatory modules governing nine gene programs, including the canonical IFN{gamma}-MHC-I axis regulating an antigen-presentation and interferon-response program. Agentic interpretation of the entire model with an AI co-scientist linked additional modules to trafficking, proteostasis and chromatin regulation. Perturb-ME, along with agentic interpretation, provide a scalable framework for comprehensive functional discovery from phenotype-enriched genetic screens.

cell biology↗

bioRxiv2026

PerturbTrace: Evaluating Feedback Use by AI Co-Scientist Agents in Perturbation Discovery

Recent advances in AI co-scientists have brought LLM agents into closed-loop experimental design. However, whether these agents use feedback from earlier rounds to revise subsequent experimental decisions remains unclear. We address this question with PerturbTrace, which evaluates each round-to-round transition through Feedback-to-State, State-to-Action, and Action-to-Outcome. These stages assess whether feedback is reflected in the agent's rationale and perturbation-selection strategy, whether the stated strategy guides the next perturbation batch, and whether that batch yields more hits than expected under random sampling. We evaluate four LLM agents on 17 screen-derived tasks and compare them with random selection, active learning, and LLM-guided Bayesian optimization baselines. Each agent outperforms the strongest non-agent method on at least 15 of the 17 tasks, yet controlled evaluations across six tasks show no consistent advantage from true feedback over random or no feedback. Among 576 transitions under true or random feedback, only 43 (7.5%) complete the full Feedback-State-Action-Outcome sequence, including 25 under random feedback. These findings show that high final recall does not necessarily indicate effective feedback use. They also highlight the need to evaluate closed-loop scientific agents by both their discovery performance and whether feedback changes their subsequent decisions.

bioinformatics↗

bioRxiv2026

Phagocytic activity of perivascular cell promotes CNS injury pathology

Injury, stroke, and neurological diseases cause persistent accumulation of cellular debris that deteriorates lesion microenvironment and impedes central nervous system (CNS) repair. Debris clearance in the injured CNS has long been attributed primarily to microglia and infiltrating macrophages. Here, we identify perivascular cells as previously unrecognized phagocytes that expand after injury and exhibit robust phagocytic activity. Perivascular cell phagocytosis is conserved across multiple mouse models of CNS injury and human stroke lesions. These cells exhibit key hallmarks of phagocytosis, including LC3-associated phagocytosis for efficient lysosomal degradation. Mechanistically, phosphatidylserine serves as the eat-me signal and Axl mediates myelin debris uptake. Myelin phagocytosis drives perivascular cell proliferation, fibrosis and lesion progression. Genetic deletion of Axl in perivascular cells or pharmacological inhibition with the FDA-approved Axl inhibitor Gilteritinib reduces pathology and improves functional recovery after spinal cord injury. Together, these findings establish Axl-dependent perivascular cell phagocytosis as a therapeutic target for CNS repair.

neuroscience↗

bioRxiv2026

Confinement considerations for CRISPR toxin-antidote gene drive systems

Gene drives are potentially powerful tools, able to spread throughout target populations. They could be used to modify or suppress disease vectors, invasive species, and agricultural pests. Yet, in many scenarios, confinement of the drive to only a target population is required. Several types of drives are capable of this, among the most promising of which are CRISPR toxin-antidote drives. Though showing good performance in simple models, such drives have not been thoroughly assessed in spatially structured populations. Here we evaluate three modification drive variants with varying levels of confinement. We find that Toxin-Antidote Recessive Embryo (TARE) drive and 2-locus TARE drives can usually spread in connected populations or from a single sufficiently large release. However, they can still be stopped by a migration corridor or by a sufficiently high population density gradient. 1-locus 2-drive TARE, on the other hand, will not be able to spread outside of release areas and indeed will often retreat in the face of wild-type alleles. When these drives arrive at a point source such as a port, only the standard TARE drive has a significant chance of establishing if releases occur at sufficiently high frequency and quantity. However, in models with parental care among a limited number of offspring, TARE drives become more invasive, with lower introduction thresholds. Overall, we find that spatial and other ecological factors can substantially affect the outcome of a confined CRISPR toxin-antidote drive release.

ecology↗

bioRxiv2026

ERK signaling integrates multiple mechanisms to drive cortical evolutionary expansion

A critical first step in human evolutionary divergence from chimpanzees and other primates is the production of greatly increased numbers of cortical pyramidal neurons during development and over evolutionary time. This expansion underpins the enlarged human cerebral cortex and, consequently, higher-order cognition and consciousness. However, the cellular and molecular mechanisms driving this progressive increase in cortical neuron number remain incompletely understood. Here we show that elevated ERK signaling in human cortical radial glia (RGs), relative to mouse RGs, arises from evolutionary changes in the developmental expression of existing shared genes. Using the Emx1-Cre line to overexpress MEK1DD, a constitutively active mutant of rat MAP2K1, we found that enhanced ERK signaling in mouse cortical RGs upregulates cAMP-PKA signaling, along with promoting self-renewal, expanding the RG pool, and accelerating cell cycle progression. Conversely, overexpression of PKA signaling in mouse cortical RGs via in utero electroporation of constitutively active PRKACA mutants (L206R or W196G) reduces ERK activity, prolongs the cell cycle, suppresses SHH-SMO and YAP/TAZ pathway activity, and blocks ependymal gliogenesis. Together, our results demonstrate that ERK and PKA signaling in cortical RGs engage a dynamic balance of synergistic and antagonistic interactions. This interplay couples evolutionarily enhanced ERK and PKA activities to increased cortical RG numbers, prolonged cell cycle progression, and an extended neurogenic period, which collectively culminate in a marked increase in neuronal production. We conclude that heightened ERK pathway activity in cortical RGs serves as a central driver of progressive neocortical expansion across mammalian evolution. Significance StatementThe human brains extraordinary cognitive abilities stem from a dramatic increase in cortical neuron numbers during evolution. This study reveals that enhanced ERK signaling in human cortical radial glia neural stem cells, relative to other primates, is a key driver of this expansion. By manipulating this pathway in mice, we demonstrate that ERK coordinates multiple mechanisms--including self-renewal, cell cycle acceleration, and a dynamic interplay with PKA signaling--to boost neuronal output. These findings identify a core molecular axis underlying mammalian cortical evolution and provide a framework for understanding how the human cortex acquired its unique size and, consequently, its remarkable complexity.

developmental biology↗

bioRxiv2026

An IL-34-IGF-1 inflammatory axis fuels KRAS-mutant lung cancer progression

Macrophages are innate immune cells of embryonic or adult origin with tissue specific roles in homeostasis, disease surveillance, and wound repair that can be co-opted to promote tumor growth and spread1-11. An understanding of the specific roles of macrophage subsets in lung tumor initiation and progression could promote new therapeutic approaches for this deadly disease. Here, we show that KRASG12D mutations in lung epithelium drive proliferation of resident, embryonically-derived alveolar macrophages, which then promote tumor cell proliferation and protection from ferroptosis, leading to tumor progression. Using genetically engineered mouse models of mutant KRASG12D non-small cell lung cancer12,13, we found that alveolar macrophages accumulate by proliferation in response to tumor cell-secreted IL-34, recapitulating events observed in late embryonic lung development. Tumor alveolar macrophages in turn drive IGF-1-dependent tumor cell proliferation. Neutralization or deletion of IL-34 suppresses IGF-1 expression, reduces macrophage and tumor cell proliferation and inhibits tumor progression. High IL34 and IGF1 correlate with poor survival in KRASG12D/V lung adenocarcinomas and in other solid tumors, indicating that bi-directional proliferative signaling between resident macrophages and tumor cells can drive human lung tumor progression. These studies identify resident macrophage-tumor cell interactions as key interception points for lung cancer therapy.

cancer biology↗

bioRxiv2026

AlfaDAX-Derived ActRIIA/B Antibody with Semaglutide Enhances Fat Loss and Improves Weight-Loss Quality in DIO Mice

GLP-1 receptor agonists achieve weight loss but are associated with clinically significant reductions in lean mass. Activin type II receptors (ActRIIA and ActRIIB) mediate signaling of myostatin and activin A, both of which negatively regulate muscle growth, suggesting that dual blockade of these receptors may preserve or increase lean mass while promoting fat loss. In this study, we developed anti-ActRIIA/B antibodies using AI-driven platforms (AlfaDAX) and selected the lead candidate AB130-165 based on in vitro binding, functional blocking, and developability assessments. Compared with a laboratory-prepared bimagrumab analog, AB130-165 exhibited potent dual inhibition of ActRIIA/B signaling, with a 9.5-fold higher functional blocking activity against activin A-induced SMAD signaling and 1054-fold improvements in binding affinity for ActRIIA (KD = 0.204 pM), 10-fold for ActRIIB (KD = 0.243 pM), respectively. In diet-induced obese mice, combination therapy with AB130-165 and semaglutide resulted in a 33.4% body weight reduction, which was superior to semaglutide monotherapy (-24.3%) and the bimagrumab combination group (-25.5%). Moreover, the combination significantly improved body composition, reducing fat mass percentage by 77.8% (vs. 65.0% in the bimagrumab combination group) and increasing the lean-to-body weight ratio to 67.3% (vs. 62.3%), demonstrating superior fat loss with better preservation of lean mass. Collectively, these findings establish AB130-165 as a differentiated anti-ActRII antibody that enables high-quality weight loss, and its combination with semaglutide shows superior efficacy over bimagrumab-based regimens. With favorable developability and potential for long-acting subcutaneous administration, AB130-165 represents a promising next-generation therapeutic candidate for obesity and muscle-sparing weight management.

pharmacology and toxicology↗

bioRxiv2026

OCellus: A Language-Model Framework for Single-Cell, Spatial, and Perturbation Biology with Natural-Language Reasoning

Computational modeling of cellular behavior--the virtual cell--has emerged as a stated grand challenge at the intersection of artificial intelligence and biology, yet existing foundation models remain specialized: single-cell models process dissociated transcriptomes only, spatial models require dedicated spatial-aware architectures, and perturbation predictors depend on manually curated knowledge bases that cap generalization. Here we introduce OCellus, a single nine-billion-parameter language model (Qwen3.5-9B) fine-tuned on twenty-two biological tasks that simultaneously addresses all three limitations through three coordinated technical contributions on a shared backbone. First, EvenClock encodes two-dimensional spatial coordinates as eighteen clockface sectors of text, enabling spatial reasoning on a vanilla language model without architectural modification; on ten spatial transcriptomics tasks OCellus attains 77 percent spatial-neighborhood accuracy, 96 percent spatial-cellchat accuracy, and 0.70 proportion-cosine similarity on spatial deconvolution, all without any spatial-aware architectural components. Second, per-gene language-model embeddings replace the Gene Ontology annotations that GEARS depends on, achieving Pearson correlation 0.945 on the Replogle 2022 perturbation benchmark versus 0.84 for GEARS across 457 completely unseen knockout genes. Third, OCellus-Agent provides a Planner-Router-Verifier natural-language interface that achieves 75 percent pipeline accuracy on eighty multi-task queries. Removing language-model embeddings collapses perturbation Pearson to 0.06, confirming that learned functional representations--not graph topology--drive the gain. As a cell-type encoder, OCellus ranks first among fourteen foundation models in linear-probe accuracy at 95.1 percent across four benchmark datasets, and reaches 72.6 percent average across twenty-two evaluated biological tasks--a 57-percentage-point absolute gain over the strongest baseline configuration. As a language model, OCellus uniquely generates natural-language explanations of its predictions, a capability absent from all competing methods. Code, pre-trained model weights, the graph-neural-network module, and the agent system will be made available upon publication.

bioinformatics↗

bioRxiv2026

PFOS aggravates atherosclerosis via Bacteroides caecimuris expansion-driven bile acid remodeling and subsequent intestinal FXR-TLR3 signaling cascade

BACKGROUNDPerfluorooctane sulfonate (PFOS) is a widely distributed persistent organic pollutant in the environment and has been associated with an increased risk of atherosclerosis. However, the underlying pathogenic mechanisms remain largely unclear. This study aimed to investigate the effects of PFOS on atherosclerosis and its associated gut-vascular axis. METHODSPseudo-germ-free mouse models and fecal microbiota transplantation (FMT) were used to determine the role of the gut microbiota in PFOS-induced atherosclerosis. Metagenomic sequencing was performed to characterize alterations in gut microbial composition following PFOS exposure, and targeted metabolomics was used to assess bile acid profiles in the ileum and plasma. Transcriptomic analysis of Bacteroides caecimuris (B.caecimuris) was conducted to explore the reasons for the increased abundance of B.caecimuris after PFOS exposure. In addition, intestinal transcriptomics and ChIP-qPCR were performed to validate transcriptional regulation within the FXR-TLR3 signaling axis. RESULTSAmong 127 participants with paired serum and fecal samples, including 82 patients undergoing coronary angiography with Gensini scores (GS score), fecal PFOS levels were significantly associated with lipid profiles and GS score, whereas serum PFOS showed no such association.Mechanistically, PFOS exposure promotes intestinal enrichment of B.caecimuris by upregulating its tolC gene, thereby enhancing efflux capacity.This microbial shift was accompanied by reduced levels of tauro-ursodeoxycholic acid (TUDCA) and aberrant activation of intestinal FXR signaling.Further analyses demonstrated that FXR activation upregulated TLR3 expression and promoted inflammatory responses and atherosclerosis progression via the TLR3-NF-{kappa}B signaling axis. Both intestinal epithelial-specific FXR deficiency (Fxr{Delta}IE) and TUDCA supplementation significantly suppressed pathway activation and alleviated disease phenotypes.Functional experiments identified TLR3 as a key downstream effector of FXR. Overexpression of TLR3 abolished the protective effects observed in Fxr{Delta}IE mice. Moreover, pharmacological inhibition of TLR3 using CU CPT-4a significantly improved established atherosclerotic lesions in vivo. CONCLUSIONSThis study identifies a gut microbiota-driven FXR-TLR3 signaling axis that mediates PFOS-induced atherosclerosis. These findings provide new mechanistic insights into environmentally induced cardiovascular disease and suggest potential targets for risk assessment and therapeutic intervention.

pathology↗

bioRxiv2026

Integrating Machine-learning and Ultra-high-throughput Screening for Enzyme spaces exploration

AbstractThe systematic navigation of biocatalyst space is constrained by elusive structure-activity rules and a lack of evolutionary history. Here, we present IMUSE, a strategy integrating machine learning with ultra-high-throughput screening. By screening millions of droplet-encapsulated de novo enzymes, we generated massive synthetic sequence-structure datasets to train models that capture their complex fitness landscapes and biophysical principles. These models effectively guide functional exploration across both sequence and novel structure spaces. IMUSE identified synergistic triple mutations yielding [~]5-fold activity improvements and discovered active second-generation designs with novel catalytic pockets, boosting the experimental success rate >4.9-fold ([~]30%). This work demonstrates how synthetic fitness landscapes bridge the data gap in de novo enzyme space, transforming stochastic search into deterministic navigation to unlock highly proficient biocatalysts beyond natural boundaries.

biochemistry↗

bioRxiv2026

A Diploid Panax Genome Reveals Ginsenoside Diversity Driven by UGT Diversification and Network Rewiring, Rather Than Gene Family Expansion

The medicinal herb Panax notoginseng produces a structurally diverse array of triterpene saponins (ginsenosides), yet the genetic basis of this chemical complexity remains unclear. Here we present a high-quality chromosome-level genome of diploid P. notoginseng and integrate comparative genomics with multi-tissue, multi-year metabolomics and transcriptomics. Surprisingly, unlike tetraploid Panax species, P. notoginseng shows no general expansion of core saponin biosynthetic gene families. Instead, lineage-specific diversification of UDP-glycosyltransferase (UGT) families, a recent burst of LTR retrotransposons, and enrichment of species-specific genes in metabolic modification pathways point to an alternative evolutionary route. Saponin accumulation follows strict spatiotemporal compartmentalisation, and co-expression network analysis reveals that the biosynthetic machinery is not static but continuously rewired during development-from a basic synthesis module in the first year to a modular pattern supporting both broad accumulation and branch-specific modification by the third year. Seventeen differentially expressed UGTs show clear tissue preferences and saponin-branch correlations. As a representative example, PnUGT33 is tightly linked to the PPD-type saponin branch; structural modelling, molecular docking and 100 ns molecular dynamics simulations demonstrate its differential recognition of diverse triterpene skeletons. Collectively, our findings establish that ginsenoside diversity in diploid P. notoginseng arises primarily from UGT lineage diversification, developmentally rewired regulatory networks and UGT mediated branch selective post-modification, rather than from expansion of core pathway genes. This work provides a new paradigm for understanding how plants achieve metabolic complexity without whole genome duplication or massive gene amplification.

genomics↗

bioRxiv2026

ContinuumCellAgent: A Framework-Guided Agent for Long-Horizon Scientific Research

AI-scientist systems are beginning to automate parts of scientific research. We present CO_SCPLOWONTINUUMC_SCPLOWCO_SCPLOWELLC_SCPLOWAO_SCPLOWGENTC_SCPLOW, an autonomous agent that executes literature review, hypothesis formation, computational experimentation, manuscript drafting, and adversarial peer review as a single unattended run. Existing AI scientist systems remain difficult to diagnose because they lack modularity, systematic prompt grounding, and observability into long-running behavior. CO_SCPLOWONTINUUMC_SCPLOWCO_SCPLOWELLC_SCPLOWAO_SCPLOWGENTC_SCPLOW addresses these gaps with a modular supernode architecture for stage-wise backend swapping, protocols grounded in curated research-method checklists that also define reviewer rubrics, and a diagnostics layer that records file-based artifacts, message traces, and state transitions. We evaluate the system on open-domain QA benchmarks and biomedical/longevity case studies, showing that it can produce checkable research artifacts while exposing pipeline dynamics for rigorous AI co-scientist research.

bioinformatics↗

bioRxiv2026

Reaction-diffusion condensation generates a regulatable landscape for self-organizing subcellular structures

Cells self-organize complex internal architectures through an interplay between biochemical signaling and physical structure. Within this paradigm, reaction-diffusion signaling mechanisms create local concentration gradients, and protein condensates assemble in a concentration-dependent manner, providing a natural coupling for a landscape of composite active structures to emerge. Here, we systematically chart this landscape in human cells at scale using RIPPLE (Reaction-diffusion IDR Platform for Producing Living Emulsions), a synthetic modular system that tethers disordered, condensate-forming IDR sequences to a programmable, two-protein wave-generating reaction-diffusion (RD) circuit. Fusing IDRs to the RD Activator, its partner ATPase, or both, generates a vast array of self-organizing subcellular architectures--ranging from traveling condensation waves to oscillating droplet networks to persistent phase-separated macrostructures that pattern the entire cell. By systematically tuning the frequency and amplitude of the underlying RD waveforms, we reveal frequency-dependent transitions regulating dilute, condensed, and aggregated states specified by IDR sequence chemistry and miscibility. RIPPLE provides a platform for understanding and engineering protein condensation far from equilibrium, revealing how signaling and structure can cooperate in a minimal, two-protein system to generate a tunable spectrum of dynamic cellular structures.

Synthetic Biology↗

bioRxiv2026

Monoclonal nephritic factors reveal insights into C3 convertase dynamics and dysregulation

C3- and C5-nephritic factors are potent but poorly understood autoantibodies that dysregulate complement convertases. To date, their underlying mechanisms of action, epitopes, and sequences remain unknown. To address this knowledge gap, we immune profiled B cells from a nephritic factor-positive C3 glomerulopathy patient and identified the first monoclonal C3- and C5-nephritic factors. We present the structure of a C3-nephritic factor bound to a C3 convertase with the convertase protease domain unexpectedly rotated and inhibited. This rotation advances our understanding of complement convertase progression and decay, explains disease-associated convertase variants, and reveals the molecular mechanism by which nephritic factors can either activate or inhibit convertase activity. We also detail heterogeneity within and between C3- and C5-nephritic factors in terms of convertase binding, stabilizing capacity, regulator inhibition, fluid-phase activation, and disease contribution. These findings improve stratification of patients with C3 glomerulopathy, redefine basic C3 convertase dynamics, and provide insights into antibody-mediated modulation of the complement system.

immunology↗

bioRxiv2026

CytoGem-XAI:A Hypergraph Neural Network Framework for Genome-Scale Metabolic Modeling and Interpretable Analysis

Genome-scale metabolic models are essential for understanding cellular metabolism, yet existing deep learning approaches remain black boxes, and traditional flux balance analysis (FBA) cannot provide sample-specific predictions. To our knowledge, CytoGem-XAI is the first framework to combine hypergraph neural network representation with interpretable, FBA-parallel analysis and sample-specific metabolic characterization. Built upon hypergraph representations where reactions are encoded as hyperedges connecting their participating metabolites, CytoGem-XAI introduces three analysis modules: perturbation-based carbon source importance ranking, hard intervention reaction bottleneck identification, and pathway-level topological attribution. Beyond prediction, CytoGem-XAI uniquely enables condition-dependent carbon source essentiality and reaction bottlenecks that vary with genetic background--capabilities absent from both traditional FBA and existing deep learning methods. Trained on 17,400 E. coli growth conditions using 10-fold cross-validation, our framework achieves R2 = 0.862, substantially outperforming AMN (R2 = 0.81, +6.4%), FBA (R2 = 0.62, +39%), and gradient boosting baselines (R2 = 0.71, +21%). Biological validation confirms that CytoGem-XAI identifies known essential carbon sources (e.g., alanine, malate) and rate-limiting enzymes (e.g., TCA cycle), while also revealing N-acetylmuramate--a peptidoglycan precursor--as a previously underappreciated essential nutrient.

bioinformatics↗