Biology subjects

Dong, Z.

Publications and source records attributed to Dong, Z..

At least 37 records · Page 2Linked to original sources

bioRxiv2026

Deep-learning-enabled morphodynamic analysis of drug responses in a biomimetic fibrin-based 3D glioblastoma invasion model

Glioblastoma (GBM) lethality arises from aggressive invasion and diffuse infiltration of brain tissue. Conventional GBM preclinical models often fail to predict clinical therapeutic efficacy because they do not recapitulate the pathological extracellular matrix (ECM) cues that drive tumor invasion. Here, we present an ECM mimetic 3D platform using a fibrin scaffold to recapitulate the hemorrhagic, pro-thrombotic tumor microenvironment characteristic of high-grade gliomas. This fibrin scaffold induces a pro-invasive phenotype in GBM spheroids by upregulating proliferation/cell cycle- (MYC, FOXOM1, CCND1) and invasion-associated-(CTSS, FOXM1, CCND1) genes. Traditional cell morphology quantification methods (e.g., circularity) distil complex shapes into singular metrics and cannot capture the nuances of invasion. To address this limitation, we have applied a deep-learning segmentation pipeline (MARS-Net) and high-content morphodynamic descriptors. By using the Preserving Heterogeneity (PHet) algorithm, the 3D platform accurately classifies invasiveness levels and captures the invasion-inhibitory effects of potential repurposable drug candidates. We demonstrate that our model can predict a spheroids long-term invasive fate with high accuracy using only partial image sets from early time-points, rather than the complete time-course images. Our work presents an in vivo-like, scalable 3D platform integrated with a quantitative high-throughput pipeline to elucidate GBM invasion mechanisms and to evaluate anti-invasive compounds.

bioengineering↗

bioRxiv2026

A cerebellar cognitive rheostat bidirectionally controls attention

Attention requires filtering distractors and amplifying signals, processes classically attributed to cortico-thalamic networks. Here, we reveal that the cerebellum operates as a bidirectional "cognitive rheostat" to optimize attentional states. In mice, the anterior and posterior cerebellar vermis exert opposing control over attention. Granule cells in the anterior vermis are functionally suppressed to gate sensorimotor noise via reticular nucleus-driven feedforward inhibition. Conversely, posterior granule cells are recruited by pontine inputs to amplify cognitive signals, a process relying on Grin1-mediated NMDA receptor plasticity. Circuit-specific manipulations targeting this push-pull mechanism, or localized pharmacological modulation, successfully rescued attentional deficits in an ADHD mouse model. These findings fundamentally expand the cerebellums role beyond motor coordination, identifying a topographic circuit algorithm essential for cognitive control.

animal behavior and cognition↗

bioRxiv2026

CellSwarm: LLM-Driven Cell Agents Recapitulate Tumor Microenvironment Dynamics and Sense Indirect Genetic Perturbations

Agent-based models of the tumor microenvironment (TME) traditionally rely on hand-coded rules that cannot generalize beyond their programmed logic. Here we present CELLSWARM, a framework that replaces rule-based cell decision-making with large language model (LLM)-driven autonomous agents. Each simulated cell maintains persistent state, 14 signal pathways, and a memory stream, with an LLM serving as its cognitive core. Using structured knowledge bases for cancer-specific context, CELLSWARM recapitulates TNBC microenvironment composition with fidelity comparable to hand-coded rules (Jensen-Shannon divergence 0.144 vs. 0.146; P=0.012 vs. random, Mann-Whitney U test). Beyond matching rule-based performance, LLM-driven agents demonstrate three capabilities absent from rule-based models: cross-cancer generalization by swapping knowledge base entries, treatment response prediction concordant with clinical data (anti-PD-1: 17.6% simulated vs. 21% clinical), and sensing of indirect genetic perturbations that propagate through intermediate signaling cascades (IFN-{gamma} KO: Agent +15.7% vs. Rules +0.3%; P=0.005). CELLSWARM demonstrates that LLM-driven cell agents can recapitulate and extend TME simulation beyond the reach of hand-coded rules.

systems biology↗

bioRxiv2026

Fully T2T pedigree assemblies reveal genetic stability and epigenetic plasticity of human centromeres across inheritance and cell-fate transitions

Centromeres are essential chromosome components yet remain poorly understood due to their highly repetitive sequence architecture. Using fully-phased telomere-to-telomere diploid assemblies from a three-generation pedigree integrated with long-read epigenomes from matched peripheral blood mononuclear cells, induced pluripotent stem cells, and neural progenitor cells, we generate allele-resolved single basepair resolution maps of centromere genetic and epigenetic dynamics across inheritance, reprogramming, and differentiation. We show that centromeric dip regions (CDRs), which define the functional core of centromeres, are positionally stable across generations and cell-fate transitions. In contrast, CDR epigenetic architecture is highly dynamic. Reprogramming markedly attenuates CDR hypomethylation, which is partially restored during differentiation in parallel with global hypomethylation of active alpha-satellite arrays and coordinated changes in nucleosome organization and protein occupancy. Centromeric remodeling is insulated from X-chromosome status, including Xa, Xi, and erosion. Finally, de novo mutations arising during reprogramming are enriched in centromeric regions but depleted within functional centromeric cores.

genomics↗

bioRxiv2026

Characterization of a GH17 laminarinase, MlGH17A, from a laminarin polysaccharide utilization locus in the marine bacterium Muricauda lutaonensis

Heterotrophic marine microorganisms have the capability to degrade and metabolize laminarin, which is the most abundant source of energy and nutrients in the marine environment, via enzymes encoded by genes clustered in polysaccharide utilization loci (PULs). In this study, a PUL potentially responsible for laminarin utilization was identified in the genome of the marine bacterium Muricauda lutaonensis strain ISCAR-4703, with conserved synteny in the genus Muricauda. A GH17 laminarinase (MlGH17A) encoded in the newly identified PUL was cloned, produced, and characterized as an endo-acting laminarinase, exhibiting the ability to degrade laminarin and laminari-oligosaccharides with a degree of polymerization (DP) greater than four into laminaribiose, laminaritriose, and laminaritetraose, with laminaritriose as the main product making up >50% of the produced oligosaccharide products. The three-dimensional model of the enzyme revealed the presence of seven putative subsites, including four glycone subsites (-4 to -1) and three aglycone subsites (+1 to +3), with a wide cleft to accommodate branches at the -2 subsite, enabling it to act on {beta}-1,3 linked backbones in polysaccharides with {beta}-1,6 linked branches. This enzyme is, along with the recently characterized {beta}-1,3 glucanosyltransglycosylase (MlGH17B), conserved in several Muricauda species and is suggested to play a crucial role in the utilization of laminarin by these bacteria. ImportanceLaminarin, a {beta}-1,3-glucan with occasional {beta}-1,6 branching, is the most abundant source of energy and nutrients in the marine environment. In this study, polysaccharide utilization loci (PULs) for laminarin degradation were identified in various marine Muricauda species, encoding a range of glycoside hydrolases and transglycosylases. In Muricauda lutaonensis ISCAR-4703, the PUL included two GH17 enzymes, separated by a GH30 enzyme and a major facilitator superfamily (MFS) transporter, a feature observed in all corresponding Muricauda PULs. A novel endoacting laminarinase from the PUL, MlGH17A, was characterized and shown to hydrolyze laminarin into laminaribiose, laminaritriose, and laminaritetraose, with laminaritriose as main product. Bioinformatic analysis showed that the enzyme lacked the typical subdomain found in GH17 plant {beta}-glucanases, leading to a lower number of aglycone subsites (+1 to +3). Instead, MlGH17A possessed more glycone subsites (-1 to -4), attributed to the {beta}3-3 loop, which was longer than in GH17 plant {beta}-glucanases.

biochemistry↗

bioRxiv2026

A family portrait of lanmodulin selectivity for enhanced rare-earth separations

Proteins offer a molecular design space to create bespoke ligands for the separation of critical metals, like rare earth elements (REs). However, data-intensive approaches to fine-tune metalloprotein selectivity are bottlenecked by the low-throughput nature of existing characterization methods. Here we invented an assay called SpyTag-Catcher Immobilization of Lanmodulin for Assaying Metal-Binding Selectivity (SpyCI-LAMBS) to measure metalloprotein selectivity en masse. This 96-format workflow was used to study the selectivity of 621 lanmodulin (LanM) orthologs for 15 REs, revealing eight distinct selectivity profiles based on sequence-to-function analyses. We discovered >200 LanMs with dampened selectivity for low-value LaIII relative to the prototypical LanM. This includes a LanM that can perform a challenging one-stage separation of PrIII from LaIII with up to >99.9 mol% purity and 83% yield. SpyCI-LAMBS is a powerful tool that can rapidly collect high-fidelity selectivity data to inform metal ion separations and machine learning-assisted metalloprotein design.

bioengineering↗

bioRxiv2026

Periplasmic gatekeeping of phage DNA entry by rSAM enzyme-mediated maturation of an effector with His-X-Ser repeats

In the relentless arms race between bacteria and phages, bacteria have evolved a variety of defense strategies to combat phage infection. However, no bacterial system has been identified that specifically target the phage DNA entry phase. Here, we present a novel bacterial antiphage system, termed HXS, which provides broad-spectrum and robust antiphage activity by interfering with phage DNA entry. The HXS system consists of two radical S-adenosylmethionine (rSAM) enzymes (HxsB and HxsC), a small protein (HxsD), and the effector HxsA with a peptidoglycan-binding domain and five His-Xaa-Ser (HXS) repeats. HxsB/HxsC catalyze rSAM enzyme-dependent maturation of HxsA, including N-terminal processing and a site-specific +8 Da modification, thereby producing a periplasmic effector required for HXS defense. Biochemical evidence supports a model in which the matured effector likely engages incoming DNA electrostatically to arrest entry, establishing an rSAM enzyme-modified protein effector in antiphage defense.

microbiology↗

bioRxiv2025

Lipoprotein lipase requires a flexible lid and stable C-terminal, both maintained by ApoC-II peptide binding

Lipoprotein lipase (LPL) is the rate-limiting enzyme responsible for hydrolyzing triglycerides in circulating lipoproteins. Reduced LPL activity contributes to hypertriglyceridemia, a major cardiovascular risk factor. LPL activity is thought to depend on the conformation of the lid domain, the lipid pore, N- and C-terminal domains (NTD, CTD), and stabilization of these domains by endogenous activators such as apolipoprotein C-II (ApoC-II). Despite major clinical significance, the structure-function relationship of LPLs functional domains and cofactors remain incompletely understood. To address this, we performed the longest known (1-{micro}s) molecular dynamics simulations of LPL independently and in complex with an ApoC-II mimetic peptide (ApoC-II-P). For the first time, we show that LPLs flexible lid can adopt multiple orientations, transitioning between open and closed states that regulate lipid pore access and catalytic activity. We also observed flipping of [~]180{degrees} by the CTD, a unique characteristic that dictates LPL activity when not in a closed lid state. Furthermore, ApoC-II-P stabilizes LPL by bridging its NTD and CTD, while maintaining an optimal lid orientation. Biochemical and cellular assays corroborate these findings, demonstrating that ApoC-II-P enhances LPL hydrolysis and supports noncanonical LPL functions. Together, these insights reveal previously unrecognized mechanisms governing LPL regulation and activity dynamics.

molecular biology↗

bioRxiv2025

Programmable conjugative CRISPR interference targeting genotoxin in the gut

Among microbially derived metabolites that influence host disease, colibactin garners increasing attention for its roles in the rising incidence of early-onset colorectal cancer. Produced by pks Escherichia coli, colibactin is a potent genotoxin, yet no approved therapeutics directly suppress it. Here, we engineered a self-transmissible conjugative plasmid to deliver CRISPR interference (CRISPRi) into multiple pks strains. This system silences transcription of colibactin biosynthetic genes and abolishes pks E. coli genotoxicity without the resistance mutations associated with wild-type Cas9-mediated bacterial inhibition. In mice, conjugation-mediated CRISPRi reduces DNA damage and pks E. coli colonization while preserving commensal diversity. Importantly, the system also lowers tumorigenesis driven by pks E. coli and outperforms a pharmacologic inhibitor in a mouse colorectal cancer model. Finally, we extend this platform to silence a second pathogenic metabolite, establishing a translational strategy to neutralize diverse microbial metabolites and expanding the toolkit for programmable live biotherapeutics in the gut.

bioengineering↗

bioRxiv2025

An Interactive AI Agent for Adaptive Modeling of RNA Region-Ligand Interactions via LLM-Generated Machine Learning Workflows

Precise modeling of RNA-ligand interactions is essential for understanding RNA functionality and designing RNA-targeted therapeutics. Current computational approaches largely focus on predicting discrete binding sites, limiting their applicability to complex RNA regions that may harbor multiple or diffuse ligand binding motifs. Here, we present RLAgent, an interactive agent framework designed to predict ligand interactions at the RNA region level, enabling higher-resolution and more flexible modeling than conventional site-centric approaches. RLAgent reframes the RNA-ligand prediction workflow as a dialogue-driven process. Through a natural language interface, users can interactively configure modeling preferences without writing code. A locally hosted large language model (LLM) acts as the core orchestration agent, automating all key components of the modeling pipeline, including data validation, feature encoding, model training, evaluation, and visualization. This agent-based design lowers technical barriers and enhances reproducibility, making RNA-ligand prediction more accessible for both computational and experimental researchers.

bioinformatics↗

bioRxiv2025

Overcoming breeding constraints in polyploid oat from evolutionary insights

Polyploidy provides adaptive advantages in plants by buffering deleterious mutations1,2. While polyploidization can enhance agronomic traits such as increased biomass, known as the gigas effect1-4, increasing genetic gain in polyploid crops remains a critical but difficult goal due to the difficulty in dissecting complex trait inheritance5,6. Here we present chromosome-scale genome assemblies for 26 Avena taxa, spanning diploid, tetraploid, and hexaploid lineages. We traced four independent polyploidization events across the genus, including the formation of A. agadiriana as an allotetraploid (AgAgAg'Ag'), and revealed a reticulate evolutionary history shaped by gene flow involving four subgenomes (A, B, C, D), for example, the hybrid speciation of A. hirtula. Transcriptomic analysis of 286 samples across 11 tissues, combined with deleterious mutation analysis from a cultivated population of 112 accessions, showed that polyploidization led to widespread functional redundancy among homoeologs, supporting a genome-wide buffering effect. However, derived allele frequency analysis revealed that, while disrupting functional genes may yield desirable traits, the buffering effect impedes the fixation of beneficial loss-of-function mutations and thereby limits breeding efficiency. Based on these integrative analyses, we propose breeding strategies to circumvent these limitations by targeting beneficial loss-of-function alleles within the complex polyploid background of oat. Our study highlights broader challenges in the improvement of polyploid crops and provides a foundation for future breeding strategies.

evolutionary biology↗

bioRxiv2025

Exploring non-invasive sexing of early chick embryos in intact eggs using Laser Speckle Contrast Imaging (LSCI) and Deep Neural Network (DNN)

The ability to image blood flow in early-stage avian embryos has significant applications in developmental biology, drug and vaccine testing, as well as determining sex differentiation. In this project, we used our recently developed laser speckle contrast imaging (LSCI) system to non-invasively image extraembryonic blood vessels and used these images to attempt early sex identification of chick embryos. Specifically, we captured images of blood vessels from 1,251 living chicken embryos between day three and day four of incubation. We then applied deep neural network (DNN) models to evaluate whether it is possible to differentiate sex based on vascular patterns. Using ResNetBiT and YOLOv5 models, our results indicate that sex differentiation from extraembryonic blood vessel images was not achievable with sufficiently high accuracy or statistical significance for practical use. Specifically, ResNetBiT had a five-fold cross-validated average accuracy of 59%{+/-}5% (fold-wise p-value, p [≤] 0.3) at day 3 and 61%{+/-}3% (fold-wise, p [≤] 0.04) at day 4. YOLOv5 had a five-fold cross-validated average accuracy of 55%{+/-}3% (fold-wise, p [≤] 0.3) at day 3 and 53%{+/-}3% (fold-wise, p [≤] 0.5) at day 4. Our findings suggest that using vascular pattern imaging alone is inconclusive for reliable early sex identification in chicken embryos.

bioengineering↗

bioRxiv2025

Neuroimaging Biomarkers of Neuroprotection: Impact of Voluntary versus Enforced Exercise in Alzheimers Disease Models

Exercise is a promising strategy for preventing or delaying Alzheimers disease (AD), yet its mechanisms remain unclear. We investigated how exercise influences brain structure, function, and behavior in a familial AD model. Mice underwent voluntary, voluntary plus enforced exercise, or remained sedentary. Neuroimaging included in vivo manganese-enhanced MRI (MEMRI). perfusion, and ex vivo diffusion MRI to assess morphometry, activity, cerebral blood flow (CBF), microstructural integrity and connectivity. Both exercise regimens induced structural and functional brain adaptations while reducing anhedonia. Voluntary exercise increased cortical and limbic volumes, particularly in the hippocampus, cingulate, and entorhinal cortex, supporting cognitive and emotional regulation. Adding enforced exercise influenced subcortical and sensory regions, including visual, motor and associative areas, supporting sensory-motor integration. MEMRI revealed increased activity in sensorimotor, limbic, and associative cortices, with voluntary exercise enhancing limbic and associative regions, and enforced exercise strengthening sensorimotor and subcortical circuits. White matter integrity improved in memory-associate pathways such as the corpus callosum, cingulum, and hippocampal commissure. Synaptic remodeling was observed in the cingulate cortex, anterior thalamic nuclei, and amygdala. Voluntary exercise enhanced CBF in the motor cortex and hippocampus, while enforced exercise limited these increases. Connectivity analyses revealed exercise-responsive networks spanning the cingulate cortex, entorhinal cortex, anterior thalamic nuclei, and basolateral amygdala, and associated tracts. Graph analyses linked running distance with increased thalamic, brainstem, and cerebellar connectivity, associating exercise intensity with plasticity. These findings highlight the ability of chronic exercise to modulate neuroimaging biomarkers through distinct but complementary pathways, reinforcing its potential as a neuroprotective intervention for AD. HighlightsO_LIExercise alters MRI biomarkers via distinct and partially overlapping mechanisms. C_LIO_LIVoluntary exercise boosts cortical and limbic regions for emotion and cognition. C_LIO_LIEnforced exercise strengthens subcortical and sensory areas for motor control. C_LIO_LIFA increases suggest memory tract reinforcement and grey matter remodelling. C_LIO_LIGraph analysis reveals plasticity in memory, emotion, and reward circuits. C_LI

neuroscience↗

bioRxiv2025

Quantifying brain-wide cerebrospinal fluid flow dynamics using slow-flow-sensitized phase-contrast MRI

Cerebrospinal fluid (CSF) flow is a key component of the brains waste clearance system. However, our understanding of CSF flow in the human brain, particularly within the brain-wide subarachnoid space (SAS), is limited due to a lack of non-invasive tools for measuring slow flow. Here, we propose a CSF flowmetry technique using phase-contrast MRI combined with a slow-flow-sensitized acquisition. It achieves high sensitivity in measuring slow CSF flow (e.g., 100 m/s), and enables quantitative measurement of the velocity and direction with whole-brain coverage, spanning from ventricles to SAS. Our proof-of-concept results demonstrate repeatable flow measurements and show that cardiac pulsation induces coherent CSF flow changes within the SAS. Our data also suggest that cardiac pulsation has a stronger driving effect on brain-wide CSF flow compared to respiration. This technique provides a valuable tool for investigating CSF dynamics and pathways to advance a holistic understanding of brain-wide CSF flow. TeaserA novel MR technique enables noninvasive, quantitative mapping of slow cerebrospinal fluid flow in the subarachnoid space across the human brain.

bioengineering↗

bioRxiv2025

Distinct Co-Methylation Patterns in African and European Populations and Their Genetic Associations

Human populations have substantial genetic diversity, but the extent of epigenetic diversity remains unclear, as population-specific DNA methylation (DNAm) has only been studied for [~]3.0% of CpGs. This study quantifies DNAm using whole-genome bisulfite sequencing (WGBS) and analyzes it alongside whole-genome genotype data to reveal a comprehensive picture of population-specific DNAm. Using a "co-methylated region" (CMR) approach, 36,657 CMRs were identified in 62 lymphoblastoid B cell line (LCL) WGBS samples, with validation in array data sets from 326 LCL samples. Between individuals of European and African ancestry, 101 CMRs exhibited population-specific DNAm patterns (Pop-CMRs), including 91 Pop-CMRs not found in previous investigations, which spanned genes (e.g., CCDC42, GYPE, MAP3K20, and OBI1) related to diseases (e.g., malaria infection and diabetes) with different prevalence and incidence rates between populations. Over half of the Pop-CMRs were asscoated with genetic variants, displaying population-specific allele frequencies and primarily mapping to genes involved in metabolic and infectious processes. Additionally, subsets of Pop-CMRs could be applicable in East Asian populations and peripheral blood-based tissues. This study provides insights into DNAm differences across the genome between populations and explores their associations with genetic variants and biological relevance, advancing our understanding of epigenetic roles in population specificity.

genetics↗

bioRxiv2025

Improved Microbubble Tracking for Super-Resolution Ultrasound Localization Microscopy using a Bi-Directional Long Short-term Memory Neural Network

Ultrasound localization microscopy (ULM) enabled high-accuracy measurements of microvessel flow beyond the resolution limit of conventional ultrasound imaging by utilizing contrast microbubbles (MBs) as point targets. Robust tracking of MBs is an essential task for fast and high-quality ULM image reconstruction. Existing MB tracking methods suffer from challenging imaging scenarios such as high-density MB distributions, fast blood flow, and complex flow dynamics. Here we present a deep learning-based MB pairing and tracking method based on a bi-directional long short-term memory neural network for ULM. The proposed method integrates multiparametric MB characteristics to facilitate more robust and accurate MB pairing and tracking. The method was validated on a simulation data set, a tissue-mimicking flow phantom, and in vivo on a mouse and rat brain.

bioengineering↗

bioRxiv2025

METTL3 Aggravates Intimal Hyperplasia by Facilitating the m6A-YTHDC1-dependent SGK1 Gene Transcription

BackgroundVascular smooth muscle cell (VSMC) migration and proliferation substantially contribute to neointimal hyperplasia related to in-stent restenosis. N6-methyladenosine (m6A) catalyzed by the methyltransferase-like 3 (METTL3)-involved methyltransferase complex is the most abundant RNA epigenetic modification in eukaryotes, but the role of m6A RNA methylation in VSMC migration and proliferation as well as neointima formation remains highly controversial. MethodsPrimary human and rat VSMCs were utilized for in vitro experiments. VSMC-specific METTL3 knockout mice (Mettl3flox/floxMyh11-CreERT2) were generated to explore wire injury in carotid arteries in vivo. Methylated RNA immunoprecipitation sequencing (MeRIP-Seq) was performed to screen for target genes by METTL3-catalyzed m6A RNA methylation. Methylation site mapping, MeRIP-quantitative PCR (MeRIP-qPCR), chromatin immunoprecipitation-qPCR (ChIP-qPCR) and reporter gene assays were applied to explore how METTL3 modulates its target gene expression. ResultsMETTL3 was consistently upregulated in the neointima from mice subjected to carotid wire injury and patients undergoing carotid endarterectomy. VSMC-specific METTL3 deficiency significantly attenuated neointima formation in carotid arteries following wire injury in mice. Accordingly, METTL3 ablation markedly repressed VSMC proliferation both in vitro and in vivo. Mechanistically, METTL3 directly catalyzed m6A methylation on serum/glucocorticoid regulated kinase 1 (SGK1) mRNA and subsequently facilitated its transcription, which relies on the established association between the SGK1 transcript and SGK1 promoter DNA by recruiting the m6A reader YTHDC1. Conversely, SGK1 overexpression abolished METTL3 deficiency-mediated suppression of VSMC proliferation and postinjury neointima formation. ConclusionsMETTL3-catalyzed m6A RNA methylation promoted VSMC proliferation and aggravated postinjury neointima formation by facilitating YTHDC1-dependent SGK1 gene transcription. Targeting the METTL3-YTHDC1-SGK1 axis to modulate VSMC proliferation may be a potential strategy for in-stent restenosis therapy.

molecular biology↗

bioRxiv2024

WMH-DualTasker: A Dual-Task Deep Learning Model with Self-supervised Consistency for Automated Segmentation and Visual Rating of White matter Hyperintensities - a Multicentre study

BackgroundWhite matter hyperintensities (WMH) are neuroimaging markers linked to an elevated risk of cognitive decline. WMH severity is typically assessed via visual rating scales and through volumetric segmentation. While visual rating scales are commonly used in clinical practice, they offer limited descriptive power. In contrast, supervised volumetric segmentation requires manually annotated masks, which is labor-intensive and challenging to scale for large studies. Therefore, our goal was to develop an automated deep learning model that can provide accurate and holistic quantification of WMH severity with minimal supervision. MethodsWe developed WMH-DualTasker, a deep learning model that simultaneously performs voxel-wise segmentation and visual rating score prediction. The model employs self-supervised learning with transformation-invariant consistency constraints, using WMH visual ratings (ARWMC scale, range 0-30) from clinical settings as the sole supervisory signal. Additionally, we assessed its clinical utility by applying it to identify individuals with mild cognitive impairment (MCI) and to predict dementia conversion. FindingsThe volumetric quantification performance of WMH-DualTasker was either superior to or on par with existing supervised methods, as demonstrated on the MICCAI-WMH dataset (N=60, Dice=0.602) and the SINGER dataset (N=64, Dice=0.608). Furthermore, the model exhibited strong agreement with clinical visual rating scales on an external dataset (SINGER, MAE=1.880, K=0.77). Importantly, WMH severity metrics derived from WMH-DualTasker improved predictive performance beyond conventional clinical features for MCI classification (AUC=0.718, p<0.001), MCI conversion prediction (AUC=0.652, p<0.001) using the ADNI dataset. InterpretationsWMH-DualTasker substantially reduces the reliance on labor-intensive manual annotations, facilitating more efficient and scalable quantification of WMH severity in large-scale population studies. This innovative approach has the potential to advance preventive and precision medicine by enhancing the assessment and management of vascular cognitive impairment associated with WMH. Code and model weights are publicly available at https://github.com/hzlab/WMH-DualTasker.

neuroscience↗