Biology subjects

Yin, Z.

Publications and source records attributed to Yin, Z..

At least 19 recordsLinked to original sources

bioRxiv2026

Interpretable Deep Learning of Endogenous Fluorophore Concentration and Entropy for Lung Tissue Classification

Fluorophore concentration measures and local histogram entropy describe different aspects of endogenous fluorophore maps. We investigate whether their combination improves discrimination between normal and cancerous lung tissue and which features contribute to predictions. Chemometric fluorescence microscopy (CFM) provides concentration maps of tryptophan, elastin, reduced nicotinamide adenine dinucleotide (NADH), and flavin adenine dinucleotide (FAD). A neural network based on generalized additive models with structured interactions (GAMI-Net) is used to compare concentration-only, entropy-only, and combined representations through explicit main effects and pairwise interactions. The mean areas under the receiver operating characteristic curve (ROC AUCs) across five-fold cross-validation were $95.9\% \pm 0.7\%$, $96.4\% \pm 1.6\%$, and $98.4\% \pm 0.7\%$, respectively; variability is reported as standard deviation (SD). Their aggregate differences are consistent with the hypothesis of complementary predictive information. FAD entropy, NADH concentration, and NADH entropy were the leading main effects in the combined model. The learned log-odds contributions include a U-shaped elastin-entropy relationship and nonmonotonic tryptophan relationships. By exposing how concentration and entropy enter predictions, this interpretable framework turns label-free fluorescence classification into specific, testable hypotheses about tissue metabolism and extracellular-matrix organization, motivating specimen-grouped and external validation.

biophysics↗

bioRxiv2026

Embryonic macrophages chronologically change their brain-entry routes and competence for neural progenitor-induced microglial differentiation

Microglia arise from macrophage precursors that enter the developing brain and acquire microglial identity, but how these processes are coordinated remains unclear. We show that despite prominent entry of macrophages from the ventricles into the mouse telencephalon at embryonic day (E) 12.5, they enter from the meninges at E11.5, preferentially in the rostral pallium. Consistent with this regional pattern, whole-mount en face imaging revealed that focal holes in the pial basement membrane were most frequently observed in the rostral pallium. We found that neural progenitors, but not endothelial cells, promote acquisition of microglial properties through a process that requires close cellular proximity and partly depends on TGF-{beta} receptor signaling. This responsiveness is substantially stronger in E13.5 than in E17.5 meningeal macrophages, indicating stage-dependent competence to acquire microglial properties. These findings suggest that microglial establishment is developmentally regulated by macrophage access to the telencephalon and responsiveness to local neural cues.

developmental biology↗

bioRxiv2026

Panvascular-JEPA: Unseen-Structure Evaluation of Factorized Predictive States for Hemodynamic Networks

Predictive representations for vascular simulation should be tested on unseen network structures and compared with the governing equations before being interpreted as general-purpose world models. We constructed a new, dimensioned but uncalibrated nodal resistance-inertance-compliance benchmark containing all 38 connected, non-isomorphic six-node graphs with five to seven edges. Twenty-four structures supplied training, validation, and new-background testing; 14 structures were wholly withheld. Across 648 parameter backgrounds and two prescribed action programs, 1,296 trajectories contained 96 forecast steps over three cycles. Four graph predictors received equal six candidate optimization searches, followed by five fixed-seed fits. Selected configurations and analysis code were locked before test trajectories were generated. Comparators included known-equation numerical integrators, an inertia-free reduction, and a five-coefficient model fitted to training trajectories. Unseen-structure mean standardized RMSE was 8.46 x 109 for four-factor OPF, 329 for single factor OPF, and 1.83 x 1031 for supervised graph prediction, compared with 1.09 x 10-4 for foursubstep known-equation integration and 6.77 x 104 for the fitted mechanism. The primary OPF-4 minus OPF-1 difference was 8.46 x 109, with a paired stratified-graph 95% percentile interval of [2.23 x 105, 2.54 x 1010]. Neural errors were strongly dependent on training seed and network structure, with extreme forecasts dominating the means. All four neural means exceeded persistence; relative performance within this comparison did not establish a usable forecasting model. The study supplies a distinct numerical test of structure transfer, factor-count control, and recursive stability. Its conclusions concern the implemented models, optimization budget, and finite graph catalogue. It establishes neither patient-specific performance nor anatomical transfer, biological remodeling, device efficacy, or clinical readiness.

bioengineering↗

bioRxiv2026

EBD-DTI: Episodic Bridge Diffusion for Zero-Shot Cold-Start Drug-Target Interaction Prediction

Predicting drug-target interactions (DTI) for entirely unseen drugs or proteins--the cold-start problem--remains a critical challenge in computational drug discovery. While sequence-based methods naturally support zero-shot generalization, they often ignore relational topology, and existing graph-based approaches either rely on global diffusion that blurs the boundary between inductive and transductive evaluation or require a few known interaction samples at test time (few-shot). We present EBD-DTI, a framework that enables zero-shot inference in graph-based DTI models without requiring any known interactions for unseen entities. The key innovation is episodic cold-start training : at each epoch, a random subset of training entities is masked and treated as pseudo-cold, forcing the model to learn cold-start inference with explicit gradient supervision. A bridge-conditioned local subgraph, together with multi-hop diffusion, provides cold entities with relational context from their nearest observed neighbors. Experiments on three benchmarks (BioSNAP, BindingDB, and DrugBank) demonstrate that EBD-DTI achieves competitive or superior performance compared to state-of-the-art methods under strict zero-shot evaluation, with episodic training improving AUC by up to 12%.

bioinformatics↗

bioRxiv2026

A structured-illumination miniscope for optically sectioned imaging and real-time neural decoding

Single-photon miniscopes enable large-scale calcium imaging in freely behaving animals but are limited by out-of-focus background fluorescence that degrades image contrast and single-cell signal fidelity. Multiphoton approaches address this limitation but remain costly and complex. Here we introduce a lightweight (<3 g), low-cost structured illumination miniscope that achieves optical sectioning in freely behaving mice. Using HiLo imaging implemented with a simple Ronchi grating and time-multiplexed excitation, the system strongly suppresses background fluorescence while preserving the speed, field of view, and accessibility of widefield miniscopes, and supports optical-sectioned multi-plane imaging to increase neuronal yield. Using hippocampal recordings, we show enhanced region-of-interest (ROI)-based signal quality and spatial information readout, allowing simple ROI-averaged signals to approach the performance of offline algorithm-extracted signals. We further demonstrate a proof-of-principle closed-loop brain-machine interface enabled by rapid online signal extraction and real-time neural decoding. Together, these results establish structured illumination as a practical and accessible strategy for achieving high-contrast calcium imaging with miniaturized microscopes.

neuroscience↗

bioRxiv2026

Rapamycin Mitigates a Sex-biased Convergent Aging Trajectory

Rapamycin extends lifespan across species, yet its cell-type-specific benefits and vulnerabilities remain unclear at whole-organism scale. Here, we present the Rapamycin Fly Cell Atlas (Rapa-FCA), a whole-organism single-nucleus transcriptomic atlas of Drosophila spanning both sexes, multiple ages, 18 cell classes, and 181 cell types. Rapamycin elicited a highly heterogeneous response, with prominent effects in reproductive, digestive, and neuromuscular systems and modest responses in most neuronal populations. Across diverse tissues, we identified a rapamycin-sensitive Convergent Aging Trajectory (CAT), marked by Fkbp12 enrichment and mTORC1-linked metabolic programs, including glycolysis and lipid synthesis. CAT-high nuclei accumulated with age and were preferentially reduced by rapamycin, especially in females, consistent with stronger female lifespan extension. By integrating CAT abundance, aging-clock predictions, and nucleus-ratio changes, we mapped sex- and cell-type-specific geroprotection effects of rapamycin. Together, the Rapa-FCA provides an organism-wide framework for resolving how rapamycin reshapes cellular aging across sex, tissue, and cellular state.

developmental biology↗

bioRxiv2026

GALILEO: Embodied AI scientist for autonomous therapeutic discovery in dynamic membrane systems

Generalizable Agentic Laboratory Intelligence for Learning, Experimentation, and Optimization (GALILEO) closes the prediction-to-intervention gap in therapeutic peptide discovery. Unlike prior AI-scientist systems, GALILEO couples an Observation-Thought-Action-Summary (OTAS) reasoning loop with robotic peptide synthesis and multimodal phenotyping. Candidate peptides are retrieved from a clinically informed peptide prior (CPP) and locally edited through auditable operations. GALILEO autonomously prioritized LRRC8C and SLC25A1 branches and used wet-lab feedback to update target beliefs, sequence policies, assay choices, and mechanism hypotheses. Peptides generated under this framework blocked LRRC8C currents, perturbed osmolyte/redox homeostasis, and promoted tumor-dependent T-cell activation, while SLC25A1 peptides disrupted citrate-export metabolism, reduced extracellular acidosis, and enhanced CD8+ T-cell function. These results establish retrieval-and-editing-based physical learning as a route for auditable, experimentally grounded interventions and transferable membrane-blocker rules.

Molecular Biology↗

bioRxiv2026

STN-DBS frequency-tuned beta subband dynamics are associated with speech tempo in Parkinson's disease

In Parkinsons disease, subthalamic nucleus deep brain stimulation (STN-DBS) frequency modulates narrow beta subbands, with low-frequency stimulation (LFS) selectively attenuating 15-23 Hz power. The negative association between this subband and speech tempo, together with the LFS-related improvement in articulatory rate, suggests that 15-23 Hz subband dynamics may serve as a candidate neural marker associated with speech tempo modulation.

neuroscience↗

bioRxiv2026

Brainana: an end-to-end preprocessing framework for macaque neuroimaging

Macaque MRI bridges non-invasive systems neuroscience with cellular and circuit-level mechanisms, but preprocessing tools remain difficult to integrate and deploy reproducibly. We present Brainana, an automated, BIDS-compatible preprocessing and visualization framework for macaque neuroimaging. Brainana integrates structural and functional preprocessing, cortical surface reconstruction, quality control, transform tracking, and atlas projection within a containerized package, with cloud access for users without local compute. Macaque-trained deep learning models support brain extraction and tissue segmentation, while image orientation standardization and macaque-specific surface reconstruction optimizations address variability across acquisitions. A viewer automatically organizes derivatives and links volumetric and surface data, enabling users to inspect anatomy, cortical measures, atlas delineations, and activity maps without neuroimaging expertise. Across 23 imaging sites, Brainana processed heterogeneous data from 130 monkeys, yielding anatomical correspondence, reliable native-space surfaces, localized task-evoked activations, and reproducible brain-wide resting-state correlations. Brainana enables reproducible, scalable, and accessible macaque MRI analysis, cross-study comparison, and multimodal integration.

neuroscience↗

bioRxiv2026

Intrinsic regulation of intestinal stem cell fate and homeostasis by Tet

How developmental progenitors navigate divergent trajectories to either establish adult stem cell pools or undergo terminal differentiation remains a fundamental gap in stem cell biology. Here, we identify the well conserved Tet protein as an essential transcriptional regulator of intestinal stem cell (ISC) establishment. Developmental Tet depletion causes region-specific ISC loss and compromises adult lifespan, while adult-specific loss drives progressive stem cell exhaustion. Overexpression of Tet leads to ISC expansion in both developing and adult guts. Utilizing a comprehensive single-nucleus transcriptomic atlas spanning gut development, we demonstrate that Tet stabilizes progenitor identity by maintaining epithelial integrity, niche signaling, and fate maintenance. By defining this developmental trajectory, we reveal Tet as a critical factor that drives proper ISC maturation and maintains long-term adult epithelial homeostasis.

developmental biology↗

bioRxiv2026

Unbiased Single-Cell Transcriptome-Proteome Co-Profiling Reveals Malignant Dormancy and Post-Transcriptional Buffering of CTCs

Leptomeningeal metastasis is driven by rare cerebrospinal fluid circulating tumor cells (CSF-CTCs). However, the mechanisms underlying their adaptation to chemotherapeutic stress remain elusive, primarily because transcriptomics alone poorly predicts functional protein states. Here, we present scMAPS, a single-cell multi-omics method that employs magnetic-assisted partitioning cell lysates to enable unbiased transcriptome-proteome co-profiling without loss-prone physical splitting and precision device. By coupling scMAPS with our custom CLEAP (CTC Label-free Enrichment and Accurate Picking) system, we performed the first deep multi-omics profiling of rare clinical CSF-CTCs before and after localized chemotherapy, detecting an average of 2,547 proteins and 7,821 genes per cell. The integrated CLEAP-scMAPS pipeline reveals a coordinated prioritizing survival over proliferation malignant dormancy phenotype and identified post-transcriptional buffering as the primary driver of treatment resistance. Our platform enables the comprehensive molecular phenotyping of rare clinical specimens, providing a highly versatile framework for decoding complex post-transcriptional regulatory networks.

cell biology↗

bioRxiv2026

Pre-trained Vision Transformers for Seizure Prediction: A Reproducible Baseline with Event-Based Evaluation and Statistical Validation

BackgroundScalp electroencephalography (EEG) based seizure prediction plays a critical role in improving the quality of life for patients with drug-resistant epilepsy, offering the potential for real-time warnings and timely interventions. Despite its clinical significance and decades of research, the field still lacks an open benchmark with reproducible baselines and deployment-oriented event-level evaluation. Most prior work relies on the small and outdated Childrens Hospital Boston (CHB-MIT) dataset and reports window-level metrics only, leaving the false-alarm burden of a real warning system underspecified. In seizure prediction, the cost of false alarm is significantly high since patients may receive painful electrical stimulation to suppress seizure. Hence, false alarms per hour (FA/h) and partial AUC (pAUC) are the most deployment-relevant metrics, reflecting alarm burden and discriminability in the low-false-alarm operating region that a usable warning system can realistically tolerate. However, few studies have systematically reported such metrics. In addition, vision transformers event-level performance under deployable FA/h constraints remains underexplored, and newer backbones such as MambaVision have yet to be evaluated under this setting. MethodsIn this work, we introduce a reproducible 5-fold benchmark derived from the Temple University Hospital EEG Seizure Corpus (TUSZ) dataset, and evaluate models using a pseudo-real-time event pipeline, reporting event-level sensitivity, false alarms per hour (FA/h) and partial AUC (pAUC). All models are compared to random predictors for statistical validation. We benchmark pre-trained vision transformers (SegFormer and MambaVision) under three EEG-to-image encoding methods, including a self-proposed Temporal-Patchify encoding for SegFormer. ResultsOur proposed Temporal-Patchify encoding method achieves state-of-the-art performance. We achieved 0.61 pAUC, which is 16.2% higher than the baseline Temporal-Tile SegFormer of Parani et al. The false-alarm burden (0.40{+/-}0.28 FA/h) is 44.4% lower than the Temporal-Tile SegFormer baseline while maintaining clinically usable sensitivity (60.7%{+/-}5.0%). We further perform statistical validation against a matched Poisson random predictor, confirming performance exceeds chance. Finally, we report end-to-end inference through-put up to 920 windows/s, confirming MambaVisions fastest inference speed, exceeding SegFormer by over 20%. ConclusionsThis work bridges the gap between seizure prediction algorithms and clinically usable seizure prediction systems in real-world settings. Our findings indicate that pre-trained vision transformers, when coupled with appropriate EEG encoding methods, can achieve robust performance in low-false-alarm operating regimes, which is critical for real-world deployment. This benchmark and evaluation framework may facilitate more clinically meaningful and reproducible seizure prediction research.

bioinformatics↗

bioRxiv2026

Enhanced intercellular transfer of mitochondria from nuclear respiratory factor 1 (NRF1)-primed mesenchymal stem cells: towards creation of superior mitochondrial delivery hubs

Mitochondrial dysfunction is a pervasive hallmark of diverse diseases. In endothelial cells (ECs), oxidative stress, bioenergetic failure, and dysregulated mitochondrial dynamics (fusion-fission, mitophagy) damage the endothelium and promote vascular pathologies such as diabetes, atherosclerosis, and aging. Mitochondrial augmentation, via direct transplantation of isolated mitochondria or cell-to-cell transfer of the organelle, has emerged as a strategy to restore mitochondrial function in metabolically compromised cells. We recently established that overexpressing nuclear respiratory factor 1 (NRF1), a driver of mitochondrial biogenesis, in mesenchymal stem cells (MSCs) increases mitochondrial content and preserves mitochondrial function under senescence-inducing stress. Here, we advance NRF1-primed MSCs as enhanced mitochondrial hubs for intercellular mitochondrial delivery to cells undergoing mitochondrial dysfunction. We hypothesized that NRF1 overexpression engages mitochondrial transfer machinery, thereby enhancing both tunneling nanotube (TNT)- and extracellular vesicle (EV)-mediated mitochondrial transfer to stressed ECs, improving EC mitochondrial fitness and health. mRNA-mediated NRF1 priming of MSCs increased expression of proteins involved in mitochondrial motility and transfer, enhanced TNT formation, and increased production of mitochondria-containing EVs. Single-cell RNA sequencing (scRNA-seq) results show that NRF1 priming shifted MSCs into distinct transcriptional states, with NRF1-enriched clusters exhibiting coordinated upregulation of cell-adhesion/cytoskeletal connectivity programs and vesicle-fusion/trafficking pathways, features consistent with enhanced structural coupling and secretory transfer capacity. NRF1 priming increased TNT-like F-actin intercellular bridges in direct co-culture and elevated mitochondria-containing EV transfer in transwell assays, demonstrating augmented mitochondrial delivery through both contact-dependent and contact-independent routes. Consequently, recipient ECs displayed reduced mitochondrial ROS, preserved membrane potential, improved oxidative phosphorylation and ATP production, rebalanced mitochondrial dynamics of fusion-fission and mitophagy. NRF1-primed MSCs further attenuated oxidative stress-induced EC senescence and apoptosis. Together, these findings identify NRF1 activation as a mechanism to reprogram MSCs into high-capacity mitochondrial donors and support NRF1-driven mitochondrial hub engineering as a strategy to strengthen mitochondrial transfer-based therapies for diseases characterized by mitochondrial dysfunction.

bioengineering↗

bioRxiv2026

The amplitude and latency of the earliest signal in V1 encode bottom-up saliency by feature conjunction

The neural origin of bottom-up saliency for exogenous attention remains highly controversial. In this study, we investigated whether the earliest activity in the primary visual cortex (V1) encodes saliency signals defined by the eye-of-origin and feature-conjunction information. Electroencephalography (EEG) recordings from the human occipital cortex revealed early responses to eye-of-origin (E) and/or orientation (O) singletons, with larger response amplitudes to the double-feature (EO) singletons. The short onset latency (58-70 ms) and polarity reversal of the responses indicate a V1 origin. Importantly, the latency and amplitude of these responses predicted behavioral detection performance. Together, these findings suggest that the timing and amplitude of the earliest signals in V1 represent the saliency of combined feature contrasts for bottom-up attention. These signals unlikely originate from projections of other proposed source areas of saliency, due to the scarcity of necessary monocular neurons to process eye-of-origin information. HighlightsO_LIEye-of-origin information, invisible to the SC, elicits an early saliency signal in V1 within 50-100 ms. C_LIO_LICombined feature contrast enhances V1 saliency responses in a nonlinear fashion. C_LIO_LIThe latency and amplitude of V1 saliency responses predict behavioral detection performance. C_LI

neuroscience↗

bioRxiv2026

DNA sequence quantitatively encodes CTCF-binding affinity at genome scale

CTCF is a central architectural protein that shapes 3D genome organization through sequence-specific DNA binding, but how DNA sequence quantitatively determines CTCF-binding strength remains poorly understood. Progress has been hampered by the lack of large, high-quality measurements of binding affinity. Here, we experimentally determine in vitro CTCF-binding affinity for 276,765 DNA sequences derived from the human genome, generating a comprehensive quantitative landscape of CTCF-DNA interactions. Leveraging these data, we develop DeepCTCF, a deep learning model that predicts CTCF-binding strength directly from DNA sequence and enables quantitative interpretation of CTCF motif grammar. Using this framework, we systematically dissect how specific sequence features modulate CTCF-binding affinity and generate quantitative predictions for disease-associated variants that alter CTCF binding. Together, this study defines general principles by which DNA sequence encodes CTCF-binding affinity and provides a quantitative framework for interpreting regulatory sequence variation. TeaserMapping how DNA encodes CTCF binding reveals quantitative rules across over 1{per thousand} of the human genome.

genomics↗

bioRxiv2025

scCotag: Diagonal integration of single-cell multi-omics data via prior-informed co-optimal transport and regularized barycentric mapping

Recent advances in high-throughput single-cell technologies have enabled characterization of cellular states across distinct omics layers, yielding complementary insights into the organization of biological systems. To jointly leverage these disparate modalities, computational methods have been developed to align non-overlapping cell populations profiled in different omics layers, a task known as diagonal integration, thereby facilitating more fine-grained biological interpretations. However, existing integration approaches unrealistically assume that all cells are alignable and treat prior-knowledge-derived feature correspondences as fully reliable, retaining them unrefined throughout the alignment process. In this work, we introduce scCotag, a co-optimal transport (COOT)-based deep learning frame-work for diagonal integration of single-cell multi-omics data, to address these limitations. scCotag first infers cell alignment and feature correspondence with prior-informed COOT in an iterative manner, and then leverages the resulting transport plans to jointly learn cell and feature embeddings via regularized barycentric mapping and graph reweighting. Systematic benchmarking on human brain, bone marrow, and blood single-cell RNA-seq and ATAC-seq datasets demonstrates the overall superior performance of scCotag over state-of-the-art methods in both cell alignment accuracy and embedding accuracy. In particular, scCotag excels in simulated imbalance scenarios where not all cells are alignable. Applying scCotag to data in postmortem brains of Alzheimers disease (AD) and non-AD (NoAD) patients further yields more fine-grained biological interpretation of regulatory mechanisms. Together, these results demonstrate that scCotag provides a robust framework for single-cell diagonal integration and regulatory inference.

bioinformatics↗

bioRxiv2025

Distributed neural dynamics underlie the shift from movement preparation to execution

Dissociating the neural mechanisms of movement preparation from those of execution across the brain is fundamental to understanding motor control. Invasive recordings in primary motor cortex (M1) indicate that shared neural populations support both phases by unfolding within orthogonal subspaces or "manifolds", yet whether and how these dynamics evolve across the broader set of brain regions involved in memory-guided skilled action remains poorly understood. Here, we used magnetoencephalography (MEG) data obtained during a memoryguided delayed finger-sequence task to track fast population dynamics across M1, premotor cortices, and the hippocampus. Linear discriminant analysis decoding of sequence-specific preparatory and execution patterns revealed a transition from preparation to execution patterns across all regions prior to movement onset. Importantly, the onset of execution exhibited region-specific timing, with M1 shifting last,[~] 100 ms before the first button press, consistent with the established cortical motor hierarchy. Low-dimensional trajectories showed that preparatory and execution states occupied distinct manifolds, yet were not fully orthogonal, suggesting partial overlap in tuning. MEG dynamics were dominated by peri-movement phase activity in the primary motor and premotor regions. Nevertheless, above-chance sequence decoding was recoverable from all motor and premotor regions during execution and in M1 during preparation. The state shift and the decoding accuracy were driven by distributed modulations across multiple frequency bands, indicating that full-band population dynamics carry richer information than band-limited features alone. These results demonstrate that non-invasive MEG can resolve continuous, hierarchical neural state transitions in the human brain in the context of memory-guided movement control. They offer mechanistic insight into distributed motor-related brain dynamics and support developing brain-computer-interfaces that incorporate signals from multiple brain regions beyond the primary motor cortex.

neuroscience↗

bioRxiv2025

ERAD Activity Distinguishes the Functional Heterogeneity of Hematopoietic Stem Cells

Hematopoietic stem cells (HSCs) rely on precisely controlled proteostasis to sustain lifelong self-renewal, yet whether intrinsic proteostatic characteristics can prospectively define functional diversity within the HSC pool remains unknown. In this study, we developed a dual-fluorescence reporter system to visualize endoplasmic reticulum-associated degradation (ERAD) activity in vivo and discovered that ERAD activity can prospectively predicts the long-term reconstitution potential of HSCs. Remarkably, ERAD states were maintained through self-renewal, revealing an inheritable "proteostatic memory" within the HSC pool. Transcriptomic and functional analyses identified SOCS2 as a key regulator of this process: SOCS2 inhibits JAK2 signaling to sustain ERAD activity, whereas hyperactive JAK2 interacts with the ERAD E3 ligase Hrd1 and suppresses VCP-mediated misfolded substrate. These findings define a SOCS2-JAK2-ERAD axis that connects cytokine signaling to intrinsic proteostasis remodeling, establishing ERAD activity as a predictive and heritable marker of HSC function and providing a conceptual framework for understanding hematopoietic stem cell heterogeneity.

cell biology↗