Biology subjects

Zhu, Y.

Publications and source records attributed to Zhu, Y..

At least 127 records · Page 7Linked to original sources

bioRxiv2026

Progressive Backmapping of Highly Coarse-Grained Protein Models

Reconstructing all-atom (AA) structures from highly coarse-grained (HCG) models remains a significant challenge in multiscale molecular dynamics (MD) simulations, particularly for mesoscale biomolecular assemblies that are beyond the reach of conventional MD methods. Building upon ProNet Backmapping, a neural-network-based thermodynamically consistent approach, we introduce a progressive backmapping framework that reconstructs AA models in a stepwise manner across neighboring resolutions, for example, from a 3-residue-per-site HCG model to a 1-residue-per-site model, then to an AA model. This progressive backmapping method achieves high accuracy across a wide range of proteins and effectively reconstructs flexible linkers in multidomain architectures. Moreover, it supports hierarchical reconstruction of complex protein assemblies, including multiple virus-like particles spanning tens of nanometers and containing hundreds of subunits. Using this framework, we demonstrate--for the first time--the ability to hierarchically backmap entire viral assemblies from HCG to full AA resolution, covering at least three different resolutions. Overall, our method provides a scalable framework for incorporating atomistic detail into mesoscale simulations of complex systems across many applications in chemistry and biology. Table of contents figure O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=70 SRC="FIGDIR/small/709104v1_ufig1.gif" ALT="Figure 1"> View larger version (45K): org.highwire.dtl.DTLVardef@4af423org.highwire.dtl.DTLVardef@e2669borg.highwire.dtl.DTLVardef@1be80eforg.highwire.dtl.DTLVardef@2e679_HPS_FORMAT_FIGEXP M_FIG C_FIG

biophysics↗

bioRxiv2026

Multisensory coding of audiovisual movies in the human hippocampus

The hippocampus receives convergent input from multiple sensory systems, yet in humans it has been studied almost exclusively through vision. Here we examine how the hippocampus contributes to sensory processing beyond the visual modality and to the multisensory integration of visual information with these other modalities. Participants were exposed to multiple repetitions of short naturalistic movie clips, each presented in four formats: auditory-only, visual-only, congruent audiovisual, and incongruent audiovisual (audio and video from different movies). Using high-resolution fMRI, we measured univariate activation and multivariate representations across the longitudinal axis and subfields of the human hippocampus. Whereas univariate analyses detected only visual responses across hippocampal subregions, with no activation for auditory stimuli and no benefit of congruent stimuli, multivariate analyses revealed robust representations of both auditory and visual scenes. The posterior hippocampus showed enhanced pattern similarity for congruent stimuli relative to unisensory stimuli, especially in the left hemisphere, demonstrating multisensory facilitation. The anterior hippocampus showed crossmodal decoding between auditory and visual versions of the same clip, driven by the right hemisphere, suggesting a more abstract representation. Finally, whole-brain searchlight analyses revealed parallel effects in cortical regions known to support multisensory integration. These findings advance understanding of auditory and multisensory coding in the human hippocampus, revealing specialization and lateralization along its longitudinal axis.

neuroscience↗

bioRxiv2026

Plant associated bacteria are a rich reservoir for multidrug efflux pumps

Plants produce antibiotic substances and bacteria need to cope with those substances to colonize plants. We analyzed the inventory of genes encoding resistance-nodulation-cell division (RND) efflux pumps originating from 282 bacterial isolates from leaves or roots of the model plant Arabidopsis thaliana. We confirmed that, on average, plant associated bacteria hold a significantly increased repertoire of genes encoding RND antiporters homologs compared to strains isolated from other ecological niches, as reported in a previous study. While some RND antiporter clades were enriched in plant colonizers, other clades found in the genomes of isolates from other environments were underrepresented. An in-depth analysis of RND antiporters from plant colonizing bacteria revealed that conserved motifs can be found in each clade, possibly contributing to substrate specificity. Interestingly, we found horizontal gene transfer markers in 10% of the antiporter homologs, suggesting that horizontal gene transfer may significantly contribute to the adaptation of bacteria to the specific chemical environment created by different organs of plant hosts. In addition, homologs from leaf-isolated bacteria showed a lower diversification, and harbored markers of horizontal gene transfer in the heavy metal exporting clade. Sequence and structural analysis revealed a high diversity in RND-antiporters, with few residues under purifying selection, indicating that RND diversity is driven by random mutations. Our findings have major implications for the origin of multidrug resistances and for our understanding of the forces shaping the outcome of plant-microbe ecology in general.

microbiology↗

bioRxiv2026

Variant-to-gene mapping identifies ARHGEF12 as a primary open-angle glaucoma effector gene

Primary open-angle glaucoma (POAG), a leading cause of irreversible blindness worldwide, has a strong genetic basis. The Primary Open-Angle African Ancestry Glaucoma Genetics study previously identified 46 risk loci associated with this disease. To pinpoint causal variants and their corresponding effector genes, we analyzed gene expression, chromatin accessibility, and conformation in two ocular cell types: trabecular meshwork cells (hTMCs) and retinal ganglion cells derived from induced pluripotent stem cells (hiPSC-RGCs). We identified 24 candidate genes in hTMCs and 56 in hiPSC-RGCs. Several prioritized genes showed altered mRNA expression, implicating pathways involved in oxidative stress, tissue homeostasis, neuronal maintenance, and retinal degeneration. Among these genes, ARHGEF12 was selected for further validation because local and distal promoter interactions nominated it in both cell types, and prior evidence reproducibly links it to POAG. While its role in hTMCs is established, its function in RGCs is unclear. hiPSC-RGCs generated from a POAG donor homozygous for the risk allele exhibited reduced ARHGEF12 expression, altered morphology, and disrupted neuronal activity. By linking POAG risk loci to candidate effector genes in ocular cell types, this study provides a framework for functional evaluation of additional POAG risk variants.

genomics↗

bioRxiv2026

How the TREX-2 complex associates with the nuclear pore

Nuclear pore complexes (NPCs) control nucleocytoplasmic transport in eukaryotes, yet their architecture remains incompletely understood. Here we report a substantially extended structure of the human NPC, obtained by combining cryo-electron tomography, crosslinking mass spectrometry, and AI-assisted integrative modeling. We resolve the molecular arrangement of TPR, NUP153, NUP50 and ZC3HC1, and reveal that five additional proteins -- TMEM209, SMPD4, GANP, Centrin-2 and ENY2 -- are incorporated into the NPC. Unexpectedly, GANP, Centrin-2 and ENY2, core members of the TREX-2 mRNA export complex, are built into the nuclear ring. This finding establishes TREX-2 not as a transiently associated factor, but as an integral NPC module, positioning it opposite of the cytoplasmic NUP214 mRNA export platform. Together, our results redefine the molecular composition of the inner ring, nuclear ring and nuclear basket. They suggest a direct structural basis that couples TREX-2-mediated mRNP remodeling to NPC-facilitated transport.

cell biology↗

bioRxiv2026

Integrated Multi-Omics Enabled by Sequential Extraction for Comprehensive Molecular Profiling of Small Extracellular Vesicles

Small extracellular vesicles (sEVs) are membrane-bound particles whose protein, lipid, and metabolite cargo reflects the molecular state of their cells of origin, making them attractive targets for biomarker discovery and therapeutic development. However, comprehensive characterization of sEVs remains challenging due to the extremely limited material available. Here, we present an integrated mass spectrometry-based multi-omics platform for simultaneous characterization of lipids, metabolites, and proteins from a single sEV sample enabled by sequential extraction, maximizing sample utilization. To enhance molecular coverage and analytical depth, the platform combines iterative tandem mass spectrometry for improved small-molecule fragmentation and nano-flow proteomics with data-independent acquisition. We achieved deep and reproducible multi-omic characterization of proteins, lipids, and metabolites using 10 million sEVs. We further demonstrated the compatibility of our multi-omics platform with sEVs isolated from plasma by ultracentrifugation, size-exclusion chromatography with ultrafiltration, and polymer precipitation, revealing purification-dependent differences in molecular profiles associated with tradeoffs in yield and purity of sEVs. By enabling integrated multi-omics from the same sample, this strategy addresses a key challenge in low-input sEV analysis and establishes a robust analytical foundation for synergistic biomarker discovery and therapeutic applications.

biochemistry↗

bioRxiv2026

ProtFlow: Flow Matching-based Protein Sequence Design with Comprehensive Protein Semantic Distribution Learning and High-quality Generation

Designing protein sequences with desired properties is a fundamental task in protein engineering. Recent advances in deep generative models have greatly accelerated this design process. However, most existing models face the issue of distribution centralization and focus on local compositional statistics of natural sequences instead of the global semantic organization of protein space, which confines their generation to specific regions of the distribution. These problems are amplified for functional proteins, whose sequence patterns strongly correlate with semantic representations and exhibit a long-tailed functional distribution, causing existing models to miss semantic regions associated with rare but essential functions. Here, we propose ProtFlow, a generative model designed for comprehensive semantic distribution learning of protein sequences, enabling high-quality sequence generation. ProtFlow employs a rectified flow matching algorithm to efficiently capture the underlying semantic distribution of the protein design manifold, and introduces a reflow technique enabling one-step sequence generation. We construct a semantic integration network to reorganize the representation space of large protein language models, facilitating stable and compact incorporation of global protein semantics. We pretrain ProtFlow on 2.6M peptide sequences and fine-tune it on antimicrobial peptides (AMPs), a representative class of therapeutic proteins exhibiting unevenly distributed activities across pathogen targets. Experiments show that ProtFlow outperforms state-of-the-art methods in generating high-quality peptides, and AMPs with desirable activity profiles across a range of pathogens, particularly against underrepresented bacterial species. These results demonstrate ProtFlows effectiveness in capturing the full training distribution and its potential as a general framework for computational protein design.

bioinformatics↗

bioRxiv2026

Non-enzymatic ABHD6 interacts with Akt-FoxO1 axis to regulate selective hepatic insulin resistance

The enzymatic function of ABHD6 on insulin secretion and insulin resistance is well documented. However, its non-enzymatic function, especially its effects on selective hepatic insulin resistance and metabolic dysfunction-associated steatotic liver disease (MASLD) is completely unexplored. ABHD6 is elevated under conditions of diet-induced obesity and aging. To define the role of ABHD6 in liver physiology, we generated liver-specific ABHD6 knockout mice, as well as liver specific overexpression of native and enzymatic inactive mutant ABHD6 mouse models. We demonstrated that ABHD6 is an unidentified regulator of selective hepatic insulin resistance and contributes to MASLD and liver fibrosis. Furthermore, we found that non-enzymatic ABHD6, rather than its enzymatic form, contributes to this regulation. Mechanistically, we found that ABHD6 translocated into the nucleus and interacted with Akt/FoxO1 axis to regulate its function. In addition, knockdown of FoxO1 in primary hepatocytes or overexpression of constitutively active mutant FoxO1 by AAV approach could completely abolish the effects of ABHD6 on glucose tolerance and gluconeogenesis. Our study reveals an entirely different mechanism underlying selective hepatic insulin resistance that involves a previously unknown non-enzymatic function of ABHD6. This study opens an avenue for the development of a novel class of ABHD6 inhibitors to treat MASLD and liver fibrosis. HighlightsO_LIABHD6 expression in the liver is increased with obesity and aging. C_LIO_LIABHD6 manipulation affects selective hepatic insulin resistance, MASLD and liver fibrosis. C_LIO_LINon-enzymatic ABHD6 interacts with Akt/FoxO1 axis to regulate FoxO1 transcriptional activity. C_LIO_LIThe effects of ABHD6 on glucose tolerance and hepatic gluconeogenesis are completely dependent on FoxO1 activity. C_LI

physiology↗

bioRxiv2026

Mitochondrial Retrograde Signaling in Arabidopsis thaliana: heterogenous, spatial and polarised aspects

Organelle retrograde signaling studies have exclusively relied on bulk tissues, obscuring cell-type specificity, spatial organization, and heterogeneity within cell types. Here, we use single-nucleus and spatial RNA sequencing to define the cellular and spatial architecture of mitochondrial stress signaling in Arabidopsis thaliana. Inhibition of the mitochondrial electron transport chain by antimycin A and myxothiazol rapidly redistributed nuclei among transcriptional states within epidermal, leaf pavement, and mesophyll lineages, revealing that mitochondrial dysfunction reshapes cell identity trajectories rather than eliciting a uniform response. Stress-associated clusters were already present at low frequency in untreated samples, but the numbers of cells increased following stress treatments, indicating that stress increases the proportion of nuclei occupying pre-existing, primed transcriptional states. Gene-level analyses revealed distinct temporal dynamics and variable cell-state penetrance of the canonical mitochondrial stress markers and identified broadly responsive genes absent from earlier marker sets. Spatial transcriptomics resolved tissue-scale responses, spanning from pan-tissue induction to cell-type- and cluster-restricted activation, and uncovered pronounced adaxial-abaxial polarity in gene expression. Cell identity and spatial inferences were supported by extensive experimental validation using a custom 465-probe 10x Xenium panel and 52 promoter-GFP reporter lines. Together, these data provide a high-resolution framework for organelle-to-nucleus signaling and a resource of cell-type markers and spatial maps to dissect mitochondrial stress signaling and its integration with developmental and environmental pathways.

plant biology↗

bioRxiv2026

Accessible and reproducible mesoscale fMRI at 5.0 T: A Pulseq-based open framework for human laminar mapping

Mesoscale, layer-specific functional MRI (fMRI) enables noninvasive access to cortical microcircuitry, yet widespread adoption has been constrained by a reliance on ultra-high field ([≥]7.0 T) systems and proprietary pulse sequences. To bridge this gap and enhance accessibility, we developed an open-source framework at 5.0 T for mapping laminar brain activity. This framework integrates a Pulseq-based 3D vascular space occupancy (VASO) sequence with an end-to-end data acquisition and analysis pipeline. At matched sub-millimeter resolution (0.8 mm in-plane), the Pulseq-based 3D implementation increased slab coverage by [~]1.82-fold and improved temporal signal-to-noise ratio by [~]1.50-fold relative to a vendor-provided 2D-VASO sequence. Validated using a finger-tapping paradigm, individual cerebral blood volume-weighted (VASO) laminar activation profiles consistently revealed the canonical "double-peak" pattern, with distinct superficial and deep peaks in the primary motor cortex. These profiles exhibited excellent cross-visit reliability (r = 0.80), and peak depths showed good spatial reliability (ICC = 0.69 for deep layers; ICC = 0.58 for superficial layers). Between-subject reproducibility was high (r = 0.86). Deploying the identical Pulseq protocol at an independent imaging site reproduced the characteristic double-peak laminar profiles (r = 0.63). At the group level, 5.0 T laminar profiles closely matched established 7.0 T findings, robustly resolving both deep and superficial peaks despite the lower field strength. Notably, for each participant, a single 13-minute VASO run was sufficient to resolve reliable laminar activation patterns that exhibited high consistency with multi-run averages (r = 0.78), highlighting the potential for high-throughput population studies or clinical research settings. The Pulseq-based 3D VASO sequence file, image reconstruction pipeline, and data analysis scripts are openly available to facilitate the adoption of this framework. This work establishes a practical route towards more accessible and reproducible mesoscale fMRI for studying human laminar functional architecture.

neuroscience↗

bioRxiv2026

Ophiuchus-Ab: A Versatile Generative Foundation Model for Advanced Antibody-Based Immunotherapy

Antibodies exhibit extraordinary specificity and diversity in antigen recognition and have become a central class of therapeutics across a wide range of diseases. Despite this clinical success, antibody design remains fundamentally challenging. Antibody function emerges from intricate and highly coupled interactions between heavy and light chains, which complicate sequence-function relationships and limit the rational design of developable antibodies. Here, we reveal that modeling antibody sequence space at the level of paired heavy and light chains is essential to faithfully capture inter-chain dependencies, enabling a deeper understanding of antibody function and facilitating antibody discovery. We present Ophiuchus-Ab, a generative foundation model pre-trained on largescale paired antibody repertoires within a diffusion language modeling framework, unifying antibody generation and representation learning in a single probabilistic formulation. This framework excels diverse antibody design tasks, including CDR infilling, antibody humanization, and light-chain pairing. Beyond generation, diffusion-based pre-training yields transferable representations that enable accurate prediction of antibody properties, including developability, binding affinity, and specificity, even in low-data regimes. Together, these results establish Ophiuchus-Ab as a versatile foundation model for modeling antibodies, providing a foundation for next-generation antibody-based immunotherapy.

bioinformatics↗

bioRxiv2026

Single-Cell Proteomics of Human Peripheral Blood Mononuclear Cells Exceeding 600 Cells per Day

Single-cell proteomic (scProteomic) measurements of peripheral blood mononuclear cells (PBMCs) are of considerable value in human health, given their involvement in the maintenance of healthy and diseased states. However, the high heterogeneity and relatively small size of immune cell types demand maximal throughput and sensitivity in proteomic measurements that have yet to be fully realized. Here, we describe an approach that addresses sensitivity and throughput through the implementation of Real-Time spectral Library Searching (RTLS), TMTpro 32-plex labelling, an updated nested-nanodroplet processing in One pot for Trace Samples (N2), and a dual-column liquid chromatography system. By prioritizing tandem mass spectrometry (MS2) features with high similarity to library spectra, RTLS enables greater identification depth and feature reproducibility than a standard shotgun MS2 approach in low-input and single-cell samples. The platform permitted 660 single PBMCs to be measured per day, with an average of 750 protein identifications per cell and 1,648 proteins in total, achieving the necessary throughput and depth to characterize immune cell populations. Application of this scProteomic method and a new cell typing informatics approach to 2,130 PBMCs enabled the identification of both major and low-frequency cell types ([~]1-2%), as well as associated proteomic markers.

immunology↗

bioRxiv2026

Associative emotional memory encoding: Insights from network stability analysis of an fMRI-driven bilinear dynamics

The interplay between emotion and memory is a central topic in cognitive neuroscience, with open questions about the underlying neuronal mechanisms. This article studies dynamic interactions among the hippocampus, amygdala, and orbitofrontal cortex during an fMRI associative memory encoding task. Participants were clustered into three condition groups: Neutral-Neutral, Neutral-Emotional, or Emotional-Emotional, and viewed image pairs associated with their assigned condition. Using the dynamic causal modeling framework, we explore several dynamic models and show that a stochastic bilinear state-space model best describes the neuronal dynamics in all conditions. Furthermore, we use graph and control theory techniques to both validate and analyze the model. In particular, we analyze the network dynamics of each condition using tools from graph theory and stability theory and discuss the differences in the strength and direction of connectivity as well as the stability of each of these networks. We confirm the prior finding that memory is enhanced in emotional conditions, in particular in the Neutral-Emotional condition. In our work, this enhanced memory is associated with increased hippocampus-amygdala coupling and overall network connectivity. In addition, we show that in the Emotional-Emotional condition, the coupling of the hippocampus and amygdala, as well as the whole network connectivity increases when the first images valence is substantially less negative rated than the second image. This pattern mirrors the Neutral-Emotional condition, where the first image is neutral compared with the second one. Moreover, our model-based analyses suggest that the amygdala predominantly influences the other two regions in the Neutral-Emotional condition, whereas the OFC plays a dominant role in the other two conditions. Combined data-driven modeling, stability analyses, and graph-theory tools led to new insights and enhanced the mechanistic understanding of cortical dynamics of emotional associative memory. We discuss these insights, utilize these analytical tools to generalize our findings to some unmeasured conditions, and highlight the potential of these techniques to inform the design of future regulatory mechanisms.

neuroscience↗

bioRxiv2026

Targeting of subcellular metabotropic glutamate receptor 5 signaling to modulate pain transmission

Metabotropic glutamate receptor 5 (mGlu5) is a class C GPCR crucial for neuronal development and synaptic transmission. mGlu5 is a potential therapeutic target in pain management and modulates pain-associated gene expression and signaling pathways. Although mGlu5 inhibitors have shown promise in treating pain, none have translated to the clinic. Up to 90% of neuronal mGlu5 expression is intracellular, although the precise locations and function of different mGlu5 intracellular pools remains unclear. Building on recent evidence showing the importance of endosome-mediated nociceptive signaling by other GPCRs, we hypothesized that endosomal pools of mGlu5 contribute to pain transmission, and that targeted inhibition of intracellular mGlu5 signaling results in superior analgesia. Using calcium mobilization assays and genetically encoded resonance energy transfer biosensors, we report that upon its activation mGlu5 recruits Gq/11 and Gs to the plasma membrane. Conversely, internalized mGlu5 in endosomes recruits only Gq/11 proteins. mGlu5 signaling is highly dependent on receptor trafficking to endosomes, with sustained nuclear ERK1/2 signaling requiring both receptor internalization and active glutamate transport into the cell. We generated pH responsive nanoparticles loaded with the mGlu5 negative allosteric modulator VU0366058 (DIPMA-VU058), enabling endosome-targeted inhibition of mGlu5. Nanoparticle encapsulation of VU0366058 enhanced inhibition of both acute and sustained nuclear ERK1/2 signaling, and significantly reduced neuronal excitability in nociceptive circuits in spinal cord slices from rats with neuropathic pain. Intrathecal administration of DIPMA-VU058 achieved superior analgesia in both inflammatory and neuropathic models of pain in mice compared to free VU0366058 and the reference compound fenobam. These studies demonstrate the importance of endosome-associated receptors for the complete mGlu5 signaling response. Furthermore, we show that manipulating the cellular distribution of an allosteric modulator can engender location-biased pharmacological effects. Together, we have revealed new and unappreciated roles for endosome-specific mGlu5 signaling and demonstrate that endosome-selective targeting may offer an alternative therapeutic approach for modulating mGlu5 activity.

pharmacology and toxicology↗

bioRxiv2026

Multimodal Behavior Scoring quantifies depression-like severity across chronic stress models and identifies stress-resilient mice

IntroductionRodent chronic-stress models are central to preclinical depression research, yet behavioral phenotypes are often assessed assay-by-assay, which can reduce cross-cohort comparability and mask heterogeneity across symptom domains (e.g., anhedonia, anxiety-like behavior, and behavioral despair). An integrative behavioral index may provide a more robust and interpretable measure of overall stress-induced severity and facilitate stratification of susceptible versus resilient individuals. MethodsWe developed the Multimodal Behavior Scoring (MBS) algorithm to integrate outcomes from standard assays--sucrose preference test (SPT), open field test (OFT), and forced swim test (FST)--into a unified severity score. MBS was evaluated in two independent paradigms, chronic unpredictable mild stress (CUMS) and chronic social defeat stress (CSDS), quantifying group discrimination, reproducibility (intraclass correlation coefficient, ICC), and phenotypic stratification. ResultsMBS robustly separated stressed from control mice (CUMS: +39.8%, p<0.0001; CSDS: +121.8%, p<0.0001) and showed high cross-cohort reproducibility (ICC>0.88). MBS further identified stress-resilient subpopulations and uncovered paradigm-specific behavioral signatures, with despair-dominant patterns in CUMS (FST-MBS, r=0.61) and anxiety-centric patterns in CSDS (OFT-MBS, r=-0.74). Excluding TST did not reduce discrimination performance. ConclusionMBS provides an integrative framework for quantifying depression-like behavioral severity and heterogeneity, enabling streamlined protocols and improving the translational utility of preclinical cohorts for biomarker discovery and antidepressant screening.

animal behavior and cognition↗

bioRxiv2026

A human-in-the-loop explanation framework for morphologically transparent AI predictions from whole-slide images

Deep learning models enable the prediction of clinical endpoints from whole-slide images (WSIs), but many such models function as "black boxes", lacking transparency about whether and which histomorphological patterns drive their predictions, hindering interpretability and clinical adoption. Here we propose a human-in-the-loop explanation framework, MorphoXAI, which provides both local and global interpretability for deep learning models by incorporating human-expert interpretations. At the global level, it reveals the histomorphological patterns on which the model consistently relies to distinguish between classes of WSIs, as well as the patterns associated with confusion between classes. At the local level, it indicates which of these patterns are used in the prediction of an individual WSI and which regions within the slide correspond to such patterns. We validated our method on a deep learning model trained for ovarian tumor histologic subtype prediction. The results show that our framework generates explanations that accurately reflect the histomorphology underlying the models predictions at both global and local levels. For interpretability and clinical utility in diagnostic contexts, human evaluation results showed that our explanations were easy to interpret, rich in diagnostic features, and directly helpful for diagnostic decision-making, thereby enhancing pathologist-AI collaboration. Our work highlights that unifying global and local explanations and grounding them in expert-interpreted morphology enhances the interpretability and verifiability of deep learning models, thereby facilitating the transparent deployment of such models in clinical practice.

pathology↗

bioRxiv2026

Functional In-Context Learning in Genomic Language Models with Nucleotide-Level Supervision and Genome Compression

AO_SCPLOWBSTRACTC_SCPLOWGenomic foundation models aim to learn general-purpose representations directly from DNA sequence, enabling sequence understanding, generation, and probabilistic reasoning across a wide range of biological tasks. Scaling such models to genomic lengths, however, remains challenging due to the tension between long-range context, nucleotide-level resolution, and practical computational efficiency. Architectural innovations have enabled increasingly long nominal inputs, but often struggle to translate additional context into meaningful performance gains, particularly in the presence of sparse functional signal along eukaryotic genomes. In this work, we revisit the design of long-context genomic foundation models from the perspective of training objective and data construction. We introduce Factorized Nucleotide Supervision (FNS), which reconciles efficient k-mer tokenization with single-nucleotide likelihoods through probability marginalization, and Genome Compression Pretraining (GCP), which reshapes the training distribution by concentrating on gene-centric and regulatory regions. Together, these techniques enable standard transformer-based models to perform functional in-context learning without sacrificing nucleotide-level fidelity or computational efficiency. Building on these ideas, we present a family of autoregressive genomic foundation models supporting contexts of up to 98k base pairs across eukaryotic and prokaryotic genomes. Across training-free evaluations and downstream fine-tuning benchmarks, our models consistently improve over prior approaches and match or exceed state-of-the-art baselines while enabling substantially more efficient inference. Together, these results demonstrate that aligning supervision and data regimes with the biological structure of genomic sequence provides a principled and effective path toward scalable and biologically faithful genomic language modeling. Models, data, and scripts for downstream analyses are publicly available at https://huggingface.co/GenerTeam.

genomics↗

bioRxiv2026

A pathogenic AMPA receptor gating mutation disrupts synapse-mitochondrion axis and stalls synapse maturation

AMPA receptors (AMPARs) are central regulators of excitatory synaptic transmission and play critical roles in activity-dependent synapse maturation and circuit development. De novo missense mutations in AMPAR subunits have been widely linked to neurodevelopmental disorders (NDDs). Despite this, how these variants lead to neuronal dysfunction remain poorly understood. Here, we investigate the consequences of a recurrent pathogenic mutation in the GluA1 subunit (GRIA1 p.A636T), which alters AMPAR gating properties and is associated with autism spectrum disorder and intellectual disability. We developed a GluA1A636T knock-in mouse model, we show that GluA1A636T mice exhibit impairments in hippocampal-dependent learning and working memory accompanied by reduced baseline activity of CA1 neurons in vivo. Mass spectrometry-based quantitative proteomic analyses of juvenile and adult hippocampal samples revealed that the A636T mutation significantly alters synaptic protein expression at both ages. Notably, the mutation drives a robust upregulation of mitochondrial proteins specifically in adult mice. Consistent with this, dendritic mitochondria in adult GluA1A636T mice exhibited altered morphology and increased oxidative stress. Electrophysiological analyses further revealed abnormalities in synaptic function, including reduced basal excitatory transmission, persistence of functionally silent synapses in adulthood, and altered synaptic plasticity consistent with impaired synapse maturation. Together, these findings demonstrate that a pathogenic AMPAR gating mutation disrupts the coordinated development of synaptic and metabolic programs in the hippocampus, linking altered excitatory signaling to delayed mitochondrial stress and enduring circuit dysfunction. Our study provides a developmental framework for understanding how disease-associated AMPAR variants impair brain function and highlights synapse-mitochondria coupling as a critical axis in glutamate receptor ionotropic (GRI) disorders.

neuroscience↗