Biology subjects

Wei, L.

Publications and source records attributed to Wei, L..

At least 37 records · Page 2Linked to original sources

bioRxiv2026

Programming Biomolecular Interactions with All-Atom Generative Model

Biomolecular interactions lie at the core of cellular life, spanning diverse molecular modalities from small molecules to nucleic acids and proteins. Nevertheless, design strategies remain separated despite shared physicochemical principles of molecular recognition. Here we present AnewOmni, a unified generative framework trained on more than 5 million biomolecular complexes, that enables transferable molecular design across molecular scales by assembling chemically meaningful building blocks at atomic resolution. We further introduce programmable graph prompts to support user-defined chemical, topological, and geometric steering during generation, exploring hybrid and unconventional chemistries beyond canonical structures. We demonstrate that transferable learning of interaction patterns and physical constraints across molecular modalities is possible, via an atom-to-block latent space capturing both atomic details and structural priors. The framework successfully designed small molecules, peptides, and nanobodies targeting the challenging KRAS G12D switch II pocket, as well as orthosteric peptides and allosteric small-molecule inhibitors for PCSK9 in the absence of known binding site, achieving 23%-75% success with only low-throughput validation, bypassing modality-specific high-throughput screening. AnewOmni is the first to succeed in functional molecular design across all scales, from small chemical entities to large biologics, and represents a stepstone towards general molecular reasoning engines, advocating a generative foundation model for biomolecular interactions to enter regimes where data and human intuition remain limited.

bioengineering↗

bioRxiv2026

Oxytocin mediates the acquisition and strategy formation of cooperation in rats

Cooperative behaviors are widespread in nature and play fundamental roles throughout animal lifespans. However, the learning processes and strategy formation underlying cooperation remain largely unknown. Oxytocin (OXT) is a neuropeptide that regulates both social and non-social behaviors, yet its roles in modulating cooperative behaviors remain to be fully elucidated. To investigate these questions, we developed a temporal coordination-based cooperative paradigm for rats and manipulated the OXT system during training. Through progressive cooperative training, rats acquired cooperation and efficiently performed the task via social interaction. This learning process was accompanied by increased OXT release in social learning-associated brain regions, whereas OXT knockout impaired learning speed. Furthermore, rats displayed different cooperative strategies, with the communication-based strategy gradually increasing and becoming predominant during the cooperative training. Both OXT knockout and inhibition of OXT-ergic neurons in the paraventricular nucleus (PVN) of the hypothalamus during learning reduced the adoption of communication-based strategy. Together, our results demonstrate that OXT is necessary for both the acquisition and the strategy formation of cooperation in rats, thereby facilitating mutual rewards. These findings provide new insights into the neural mechanisms by which OXT regulates social behaviors and offer potential therapeutic implications for neuropsychiatric disorders involving social deficits, such as autism spectrum disorder and social anxiety disorder.

neuroscience↗

bioRxiv2026

Autoregressive forecasting of future single-cell state transitions

Predicting future cell state transitions from static transcriptomic snapshots is vital for understanding and steering cellular dynamics. Although existing trajectory and RNA velocity methods can infer local transitions within observed data, they are fundamentally limited in extrapolating long-range cellular dynamics. Here, we introduce CellTempo, an autoregressive generative framework that enables long-range forecasting of cell state trajectories directly from static transcriptomic snapshots. CellTempo is pretrained on scBaseTraj, a large-scale collection of multi-step cellular transition sequences constructed by integrating diverse experimental technologies and data sources. Across diverse biological systems, CellTempo accurately predicts long-range lineage progression and reconstructs unseen cellular potential landscapes. It further distinguishes transient perturbation responses from long-term cell-fate changes, enabling prioritization of candidate compounds for cell-fate engineering. These results demonstrate that long-range cellular dynamics are inferable from static observations, establishing a foundation for predictive and controllable modeling of cell fate.

bioinformatics↗

bioRxiv2026

Optical metabolic imaging of the tricarboxylic acid cycle

The tricarboxylic acid (TCA) cycle lies at the core of cellular metabolism, integrating energy production, biosynthesis and redox homeostasis, yet direct quantitative imaging of its activity in living systems remains challenging. Here we introduce MATRIX-SRS (Metabolic Activity TRacing of the trIcarboXylic acid cycle by Stimulated Raman Scattering microscopy), a platform enabling spatially resolved quantification of TCA-linked metabolism in live cells. Using emerging deuterium-labeled probes, MATRIX-SRS visualizes subcellular TCA-associated carbon-deuterium bonds in live cancer cells and neurons. We then integrate density functional theory, reaction network mapping, and hyperspectral MATRIX-SRS to construct a robust in situ metabolic quantification pipeline. Integrating MATRIX-SRS with isotope-tracing mass spectrometry, we reveal a global attenuation of TCA activity during epithelial-to-mesenchymal transition, providing deep molecular insights. Applying this framework, we further quantify changes in deuterium-labeled biomass in absolute concentrations for the first time, under native and drug-treated conditions, establishing a generalizable foundation for live quantitative spatial metabolomics.

biochemistry↗

bioRxiv2026

How different AI models understand cells differently

AI single-cell foundation models (scFMs) are believed to be able to learn essential relations in cell transcriptomics with the attention modules in Transformer, but there is no method to reveal what they actually learned. We observed that different models may grasp different aspects of relations. To unravel the mystery, we propose scGeneLens, a framework for dissecting how scFMs perceive cells. We employed a sparse block attention to replace the original attention mechanism to concentrate attentions into a few dominant gene-gene relations, used attention propagation to trace how the relations propagate across Transformer layers, and used integrated gradients to disentangle the relative contributions of gene identity and expression in cell representations. We applied it to scFoundation and scGPT and show that they exhibit pluralistic perceptions of cells: scFoundation emphasizes relations among cell-type marker genes, resulting in stronger cell-type separability, whereas scGPT focuses more on genes involved in shared cellular pathways and core biological activities, leading to representations that generalize across conditions. The framework provides a unified lens for probing what scFMs learn about cells and offers actionable insights for the design of future cellular foundation models. Our code can be seen in https://anonymous.4open.science/r/scGeneLens-B771/.

bioinformatics↗

bioRxiv2026

The mood stabilizer lithium alters behaviour and physiology via the gut brain axis.

Lithium, introduced 75 years ago by John Cade1, remains the most effective mood stabilizer for bipolar disorder2. Lithium is proposed to modulate an array of cellular pathways, many ubiquitous to all cells, with pleiotropic roles unlinked to bipolar disorder or lithium responsiveness in genome-wide association studies3,4. These mechanisms cannot explain lithiums specific effects on mood and behaviour. We demonstrate that lithiums primary action is in the periphery, not in the brain itself. Lithium acts in the gut to trigger behavioural and physiological changes, akin to those associated with a torpor-like state, that protect individuals from ingested toxins. Lithium activates gastrointestinal enterochromaffin (EC) cells via their Trpm2 cation channels to modulate afferent vagal and area postrema inputs to the brain. Eliminating these inputs by focal brain lesions eliminates lithiums effects, as does ablation of EC cells or their Trpm2 expression. Lithiums Trpm2-dependent activation of EC cells also occurs in human gut tissue, providing translational relevance for our discovery. These findings challenge the prevailing perception that lithium acts directly on the brain. Via a previously unsuspected gut-brain pathway, lithium engages brain circuitry that reduces arousal and interaction with the external world, therapeutic goals in the manic phase of bipolar disorder.

neuroscience↗

bioRxiv2026

Abalone genomics reveals an ancient asymmetry axis and the multifunctionality of the mantle

Gastropod diversification represents one of the most spectacular evolutionary radiations, underpinned by a fundamentally asymmetric body plan and diverse pigmentation patterns. To understand the genetic underpinnings of these cardinal traits, we constructed a chromosome-level genome assembly for the rainbow abalone, Haliotis iris. Integrating systematic genomics, histoembryology, and molecular assays, we uncover an ancient regulatory axis, conserved for over 544 million years, in which the long non-coding RNA lncRNA1 functions as a competing endogenous RNA to shield pitx transcripts from miRNA-mediated degradation, thereby driving the establishment of asymmetric body patterning, including the mantle. Furthermore, we identify a wnt-mitf-tyr cascade as the driver of melanogenesis and demonstrate that the mantle is the primary site of melanin synthesis. Concurrently, we reveal the mantle as a multifunctional organ possessing mechanosensory capabilities. These findings elucidate the ancient genetic toolkit governing trait evolution in gastropods. TeaserA deeply conserved RNA switch directs gastropod body plans and reveals the mantles hidden talents.

genomics↗

bioRxiv2025

A Machine Learning- 3D Microvessel Platform Identifies Kinase Targets Restoring Blood-Brain-Barrier Endothelial Integrity.

Disruption of the brain endothelial barrier is a hallmark of traumatic brain injury (TBI), and contributes to cerebral edema, coagulopathy, and delayed neurological deficits, yet there are no mechanism-guided therapies that directly stabilize human brain vessels. Here we used a three-dimensional (3D) human brain endothelial vessel platform with quantifiable macromolecular barrier function to assess responses to thrombin and plasma from patients with TBI. This endothelial-focused 3D system isolates the central cellular regulator of blood-brain barrier function and enables mechanistic analysis of endothelial signaling and drug responses as a foundation for personalized vascular therapeutics. By integrating this platform with machine learning-guided kinase analysis, we identified kinase pathways associated with barrier stabilization versus disruption and nominated inhibitors that reversed thrombin- and TBI plasma-induced leakage as well as compounds that exacerbated endothelial loss. This work establishes an important first step toward precision vascular therapeutics in TBI, by linking patient-derived plasma phenotypes to specific endothelial barrier responses and pharmacologically tractable kinase signaling, and it provides a generalizable framework for mechanistic drug discovery across neurovascular disorders. SIGNIFICANCE STATEMENTBlood-brain barrier (BBB) disruption drives cerebral edema, coagulopathy, ischemia, and neurological deficits after TBI. Targeting kinases as therapeutics is promising but challenged by complex inhibitor profiles and patient variability. We present a machine learning-guided platform built on 3D perfusable human brain endothelial-centric microvessels to systematically screen and select kinase inhibitors that reverse TBI-induced endothelial barrier breakdown. By pinpointing key kinase and gene regulators and demonstrating restoration of endothelial barrier integrity in patient-derived plasma models, our approach provides a transformative strategy for stabilizing the endothelial barrier after TBI and offers a versatile framework for therapeutic discovery across neurovascular disorders.

bioengineering↗

bioRxiv2025

Melody: Decoding the Sequence Determinants of Locus-Specific DNA Methylation Across Human Tissues

DNA methylation is a fundamental epigenetic modification that plays crucial roles in transcriptional regulation, cellular differentiation, and genome stability. However, how locus-specific DNA methylation is determined by intrinsic DNA sequence remains poorly understood. Here, we introduce Melody, a deep learning framework that predicts DNA methylation from 10-kb genomic sequences, enabling the integration of both local and long-range sequence signals. Across 39 human tissues, Melody accurately predicts methylation profiles and consistently outperforms existing state-of- the-art methods in whole-chromosome, hypomethylated-region, and cell-type-specific benchmarks. Melody also generalizes to methylation quantitative trait locus (meQTL) effect prediction and identifies regulatory sequence motifs associated with methylation variability. To extend prediction beyond profiled tissues, we further develop Melody-G, which incorporates single-cell RNA-seq foundation model embeddings to infer methylation states in previously unseen cell types directly from transcriptomic data. Together, Melody provides a unified framework for linking genomic sequence and cellular state to DNA methylation and offers new insights into the regulatory logic governing the human methylome.

bioinformatics↗

bioRxiv2025

LABEL-FREE HEMODYNAMIC MAPPING VIA ULTRAFAST POWER DOPPLER-ENHANCED PHOTOACOUSTIC IMAGING REVEALS EARLY THERAPY-INDUCED FUNCTIONAL CHANGES IN THE TUMOR.

Real-time, non-invasive imaging of tumor vascular function is critical for assessing treatment response and informing therapeutic decisions. In this study, we present a dual-modality imaging approach that integrates Ultrafast power Doppler ultrasound (UPD) with multispectral photoacoustic (PA) imaging to quantify spatiotemporal changes in tumor oxygenation and perfusion following photodynamic therapy (PDT). Our results demonstrate that PDT induces dose- and time-dependent spatial heterogenic alterations in both oxygenation and perfusion. To better interpret the spatial interplay between tumor oxygenation and vascular perfusion, we developed a Vascular Function Index (VFI), which classifies tumor subregions into four functional quadrants based on perfusion and oxygenation status at every voxel within the 3D tumor region. This framework revealed dynamic, time-dependent changes in tumor hypoxia and perfusion, with high-dose treatment groups showing a progressive increase in hypoxic and non-perfused regions, underscoring the spatiotemporal nature of PDT-induced vascular response. TeaserTumoral Blood flow and oxygenation interplay, assessed by label-free imaging, unravels heterogeneous therapy response.

biophysics↗

bioRxiv2025

Deep Learning-Augmented Stimulated Raman Imaging for Cell-Type-Specific Metabolic Profiling in Live Neuronal Co-Cultures

Neuronal metabolism is fundamental to brain functions and diseases, yet its spatial and temporal dynamics and interactions remain poorly understood. Here, we introduce a tandem deep-learning approach integrated with bioorthogonal chemical imaging using stimulated Raman scattering (SRS) microscopy. This method achieves high-speed and quantitative metabolic profiling in live neuronal co-cultures. Our deep-learning framework consists of a recurrent convolutional neural network (RCNN) that enables high-resolution 3D imaging with minimal photodamage and a U-Net segmentation model for cell-type-specific metabolic analysis. Using deuterium-labeled metabolites, we demonstrate the ability to trace lipid, protein, glucose, and D2O metabolism in neurons, astrocytes, and oligodendrocytes under physiological and pathological conditions, including NMDA receptor activation, proteasome inhibition, and Huntingtons disease. Our findings reveal distinct metabolic adaptations among neuronal cell types and underscore the importance of non-invasive metabolic profiling for understanding neuronal interactions and disease mechanisms. This platform significantly advances live-cell dynamic imaging with broad applications in neuroscience, disease modeling, and therapeutic screening.

neuroscience↗

bioRxiv2025

Microstructural functional coupling as a multimodal biomarker of emotional state transitions in bipolar disorder

Bipolar disorder (BD) is characterized by dynamic transitions between depressive and manic states, yet the neural mechanisms underlying these state shifts remain unclear. Here, we developed an integrative framework to quantify the microstructural-functional coupling (MFC) of the brain, which captures the distributional similarity between voxel-wise diffusion and functional features within each cortical region. Using multimodal MRI data from 72 BD patients and 65 matched healthy controls, we observed a global increase of MFC of BD, which indicates a de-similarity of brain microstructure and function. Partial least squares correlation (PLSC) analysis revealed that the principal latent component linking MFC to behavioral scores explained 67.2% of the variance, with the strongest contributions from the control and somatosensory networks. Mediation analysis demonstrated that regional MFC influenced anxiety indirectly through depressive symptoms, supporting a sequential affective transition from depression to anxiety. Moreover, macroscale gradient analysis showed that the principal MFC gradient was significantly correlated with neurotransmitter receptor distributions, including 4{beta}2, 5-HT1B, and H3. Together, these findings highlight MFC as a multimodal biomarker sensitive to emotional state transitions in BD and suggest its potential as a bridge between microstructural alterations, functional dynamics, and neurotransmitter systems.

neuroscience↗

bioRxiv2025

Unified modeling of cellular responses to diverse perturbation types

Cells respond to diverse external interventions through shared regulatory mechanisms, suggesting that diverse interventions may be amenable to unified computational modeling. In practice, however, these responses are profiled across highly heterogeneous experimental settings, including distinct perturbation modalities, dosages, combinations, and cellular contexts. As a result, most computational models remain narrowly tailored to a single perturbation modality or experimental setting. They are difficult to extend to new cell types or perturbation types for which limited training data are available, and offer limited capacity to reuse information across heterogeneous perturbation datasets. Here, we introduce X-Pert, a general perturbation modeling framework that jointly represents external interventions and cellular contexts within a unified multimodal architecture. X-Pert adopts a mechanism-aligned design that explicitly models gene-perturbation interactions together with gene-gene dependencies through dedicated attention mechanisms, enabling the unified handling of heterogeneous experimental settings. Across benchmarks involving both genetic and chemical perturbations, X-Pert consistently outperforms existing methods under conventional accuracy metrics as well as biology-aware evaluations. Importantly, X-Pert exhibits strong generalization across cell types and supports joint learning across perturbation types. The unified latent space learned by X-Pert further enables downstream analyses such as perturbation retrieval and drug-gene association discovery, facilitating the prioritization of candidate gene inhibitors and the identification of anti-cancer compounds. Together, X-Pert establishes a versatile and generalizable foundation for perturbation modeling and predictive virtual cells, with broad applications in biomedical research and therapeutic discovery.

bioinformatics↗

bioRxiv2025

CellTok: Early-Fusion Multimodal Large Language Model for Single-Cell Transcriptomics via Tokenization

Large language models (LLMs) can process diverse forms of information once they are represented as tokens in a shared sequence space. However, single-cell transcriptomes remain a foreign modality to LLMs because they are continuous, high-dimensional molecular profiles rather than discrete linguistic units. Here, we propose CellTok, a tokenized single-cell language modeling approach that converts transcriptomic profiles into compact cellular token sequences and incorporates them into the vocabulary of a pretrained LLM. By representing cells as native tokens, CellTok enables cellular measurements, textual instructions, biological context, and multi-cell populations to be jointly processed within the same autoregressive modeling framework. Across diverse tasks, CellTok enable LLMs to recognize individual cells, interpret homogeneous and heterogeneous cell populations, infer disease-associated cellular states, predict cell-cell communication, model developmental trajectories, and generate cellular states. Moreover, prompt-based experiments show that providing appropriate biological context improves performance, indicating that CellTok can leverage LLM knowledge and contextual reasoning to support cellular data interpretation. These results demonstrate that single-cell transcriptomes can be transformed from a foreign molecular modality into a native language for LLMs, establishing a unified interface for modeling cells, populations, and biological knowledge in a shared token space.

bioinformatics↗

bioRxiv2025

Scaling Personalized TMS: A Scalp-Based MRI-guided Alternative to Neuronavigation

BackgroundPersonalized transcranial magnetic stimulation (TMS) has shown promise in treating psychiatric and neurological disorders, but its accessibility is limited by reliance on costly and complex neuronavigation systems, particularly in resource-constrained clinical settings. ObjectiveTo develop and validate a scalp-based alternative to neuronavigation for personalized TMS targeting. MethodsWe developed a Continuous Proportional Coordinate (CPC) system-based approach for personalized TMS targeting and validated its performance through three operators across ten participants, using individualized DLPFC-SGC targets as an example. The validation framework included targeting consistency assessment, reproducibility evaluation, and cross-cortical generalizability analysis. ResultsThe CPC-based approach demonstrated comparable performance to neuronavigation, achieving 99.4% of its electric field strength (74.80 vs. 75.14 V/m; r = 0.962) and 93.9% of DLPFC-SGC functional connectivity (-0.216 vs. -0.230; r = 0.949), with high inter-operator reliability (ICC [≥] 0.903). Reproducibility analysis showed strong correlation between sessions (r = 0.948-0.957) with high reliability. Further simulations across 1,125 uniformly distributed cortical targets supported generalizability of this approach, with stable electric field preservation (99.07 {+/-} 1.54%) across different brain regions. ConclusionsThe CPC-based approach offers a practical alternative to neuronavigation, maintaining targeting accuracy and providing a more accessible option for personalized TMS implementation.

neuroscience↗

bioRxiv2025

Prior photon measurement density function constructed in a standard channel space for functional near-infrared spectroscopy

Photon measurement density function (PMDF) determines the measurement locations and their associated sensitivity to the hemodynamic fluctuations of an fNIRS channel, which is essential for pre-experiment optode arrangement and post-experiment signal analysis/results interpretation in fNIRS studies. However, obtaining the exact PMDF relies on individual MR images, which are unavailable in common fNIRS studies. Previous studies alternatively adopt the PMDF derived from single head templates, i.e., template-PMDF, which ignores the anatomical difference between the template and the participants, as well as the anatomical variability across participants. Here, we propose a methodological framework using a standard channel space to construct a prior-PMDF from a structural MRI database for arbitrarily given channel locations. We demonstrate the utility of the prior-PMDF in estimating channel-wise fNIRS imaging parameters, such as subject-specific PMDF, sensitivity in intracerebral and extracerebral tissue compartments, and sensitivity to regions defined by brain atlas. The performance of the prior-PMDF tested on an independent dataset was compared with the template-PMDF. The results show that the prior-PMDF significantly surpasses the template-PMDF.

bioengineering↗

bioRxiv2025

Lineage-Specific Epigenomic Mapping of Proximal and Distal Fallopian Tube Epithelium Reveals Mesenchymal/Wnt-Driven Tumor Phenotype Diversification

High-grade serous ovarian cancer (HGSC) typically originates from the distal fallopian tube epithelium (disFTE), while the proximal epithelium (proFTE) remains less explored. Here, Using isogenic murine organoid models with common oncogenic drivers (Tp53-/-; Krasmut; Pten-/-), we show that disFTE and proFTE generate distinct tumor subtypes driven by region-specific transcriptional programs. Unlike disFTE, proFTE demonstrates continuous activation of epithelial-mesenchymal transition (EMT) across wild-type, mutant organoids, and derived tumors. Additionally, mutated proFTE organoids and their tumors exhibit persistent WNT signaling activation and grow independently of WNT signals. Chromatin profiling shows that mutant proFTE maintains a distinctive epigenetic profile with promoters associated with EMT and WNT pathways. Human proFTE with BRCAmut also shows epigenetic activation of EMT- and WNT-related genes. Tumors near the uterotubal junction (UTJ)--anatomically akin to proFTE--exhibit a pronounced EMT signature, increased Vimentin+E-cadherin+ and nuclear {beta}-catenin expression. These findings suggest proFTE-derived tumors may occupy a unique niche in gynecological malignancies, with regional identity shaping tumor phenotypes.

cancer biology↗

bioRxiv2025

NSF is required for diverse endocytic modes by promoting fusion and fission pore closure in secretory cells

The ATPase N-ethylmaleimide-sensitive factor (NSF), known for disassembling SNARE complexes, plays key roles in neurotransmitter release, neurotransmitter (AMPA, GABA, dopamine) receptor trafficking, and synaptic plasticity, and its dysfunction or mutation is linked to neurological disorders. These roles are largely attributed to SNARE-mediated exocytosis. Here, we reveal a previously unrecognized role for NSF: mediating diverse modes of endocytosis--including slow, fast, ultrafast, overshoot, and bulk--by driving closure of both fusion and fission pores. This function was consistently observed across large calyx nerve terminals, small hippocampal boutons, and chromaffin cells using capacitance recordings, synaptopHluorin imaging, electron microscopy, and multi-color pore-closure imaging. Results were robust across four NSF inhibitors, gene knockout, knockdown, and specific mutations. These findings establish NSF as a central regulator of membrane fission, kiss-and-run fusion, endocytosis, and exo-endocytosis coupling--offering new mechanistic insights into its diverse physiological and pathological roles in synaptic transmission, receptor trafficking, and neurological diseases.

neuroscience↗