Biology subjects

Ding, R.

Publications and source records attributed to Ding, R..

At least 19 recordsLinked to original sources

bioRxiv2026

Differential subcortical-cortical circuit reorganization in autism and ADHD

Autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD) often co-occur and share symptoms as well as genetic risk factors. While subcortical-cortical networks have been implicated in both conditions, particularly involving thalamic and striatal circuits, patterns of shared and diverging disruptions remain unclear. Here we integrated resting-state functional connectivity (FC) and regression dynamic causal modeling (rDCM) across a transdiagnostic multi-site cohort. We used statistical group comparisons and individualized normative deviation mapping to map network deviations in ASD and ADHD relative to neurotypical controls and against one another. Multivariate associative techniques examined associations to clinical and cognitive phenotypes. Our transdiagnostic analyses revealed significant differentiation of ASD and ADHD across subcortical-cortical circuits. While ASD was characterized by widespread thalamocortical hyperconnectivity and bidirectional elevated coupling across thalamic subdivisions, ADHD exhibited a focal pattern of striatocortical disruption, anchored predominantly within the caudate. Multivariate analyses identified shared and disorder-specific brain-behavior associations. Across both conditions, aberrant striatocortical signaling was associated with reduced overall intelligence measures. Clinically, ASD-related thalamocortical connectivity disruptions mapped onto generalized cognitive impairment, whereas ADHD-related subcortical alterations tracked hyperactive and inattentive symptom dimensions. Our findings differentiate ASD and ADHD through distinct patterns of functional and directional subcortical-cortical dysregulations.

neuroscience↗

bioRxiv2026

Functional reorganization of large-scale brain systems underlies emotion regulation maturation

Emotion regulation via cognitive reappraisal is essential for socioemotional development and mental health, yet the organizational logic of its supporting brain systems across youth remains unclear. Using a large multisite cohort, we demonstrate that maturation of reappraisal capacity reflects reorganization of a distributed architecture rather than progressive strengthening of prefrontal control. Reappraisal robustly reduced negative affect, with non-linear gains concentrated between 6 and 10.5 years. Bayesian systems identification resolved this architecture into four functionally distinct systems-reappraisal-specific, common-appraisal, modifiable-emotion, and non-modifiable-emotion components-spanning prefrontal-control, salience, default-mode and dorsal-attention networks. Age-related variation in anterior temporal cortex, emerged as essential hub, statistically accounted for developmental gains in reappraisal success. Multivariate modelling together with intersubject analyses further identified a cross-site-generalizable whole-brain reappraisal signature and revealed an age-dependent shift in brain-behaviour coupling, with neurobehavioural convergence increasing with age. These findings establish distributed network reorganization as a core mechanism of emotion regulation development and provide mechanistic targets for stage-matched interventions.

neuroscience↗

bioRxiv2026

FUNCTIONAL FRAGMENTATION AND STRUCTURAL DRIVERS OF THALAMO-CORTICAL CIRCUITS IN TEMPORAL LOBE EPILEPSY

AO_SCPLOWBSTRACTC_SCPLOWO_ST_ABSObjectiveC_ST_ABSIn temporal lobe epilepsy (TLE), the thalamus acts as a nexus in a pathophysiological network that implicates mesiotemporal, subcortical, and neocortical regions. Studying a large multimodal and multicentre dataset, we profiled thalamic, hippocampal, and neocortical functional connectivity (FC), assessed structural mediators, and examined clinical associations. MethodsWe studied resting-state FC alongside structural and diffusion MRI data in 250 unilateral TLE patients and 259 healthy controls, with measures aggregated across four independent datasets. Data were processed using open-access neuroinformatics workflows and analyzed at a subregional level to maximize anatomical precision. Statistical analysis and mediation models assessed between-group FC changes, structural contributors, and clinical correlations. ResultsCompared to controls, TLE patients presented with reduced thalamo-cortical FC, which was most marked in mesiotemporal, fronto-central, and occipital regions. Thalamo-hippocampal FC was also reduced, with effects seen in all CA subfields. In the thalamus, FC reductions peaked in the ventral posterior nucleus when considering neocortical target regions and in the mediodorsal nucleus when considering hippocampal target regions. While ipsilateral hippocampal volume and diffusion changes mediated thalamo-hippocampal FC, thalamo-cortical FC appeared decoupled from structural alterations. Findings were consistent in left and right TLE patients, in patients with short and long disease duration, and across imaging sites, suggesting that thalamo-cortical FC imbalances are a consistent signature of TLE. Conversely, thalamo-hippocampal FC was elevated in patients with focal-to-bilateral-tonic-clonic seizures and FC alterations were more marked in the subgroup of operated patients that became seizure-free after surgery. ConclusionOur multi-site findings demonstrate marked thalamic circuit fragmentation in TLE. Ipsilateral findings robustly showed subdivision-specific effects, which point to both mesiotemporal co-lateralization as well as broader system-level involvement. Mediation analyses furthermore confirmed a key role of hippocampal pathology in disrupted thalamo-hippocampal connectivity in TLE, while broader thalamo-cortical fragmentation becomes increasingly independent of mesiotemporal compromise. Critically, thalamic FC represents a network substrate for seizure generalization and can serve as a prognostic indicator for surgical outcome. These results underscore the contribution of the thalamus as a hub in macroscale dysfunction in TLE.

neuroscience↗

bioRxiv2026

MICAFlow: Fast and Robust MRI Preprocessing Bridging Research Neuroimaging and Clinical Practice

MICAFlow is a fully automated MRI preprocessing pipeline designed to translate advanced neuroimaging workflows from research into routine clinical practice. The pipeline emphasizes speed, robustness, and ease of use, focusing on structural and diffusion MRI. Key innovations include a Label-Augmented Modality-Agnostic Registration (LAMAReg) technique driven by deep learning segmentations for reliable cross-modal alignment, integration of state-of-the-art distortion corrections, and adherence to reproducible standards (Snakemake workflow, BIDSApp specifications). We describe the design of MICAFlow and evaluate its performance across heterogeneous datasets. First, accessibility: MICAFlow processes a multimodal MRI exam in minutes with clinically accessible hardware and without requiring GPU access, making it feasible for same-day clinical use. Second, registration accuracy: LAMAReg achieves cutting-edge multi-modal registration accuracy, yielding accurate alignment of diffusion MRI, FLAIR, and intra-subject T1-weighted images while remaining generally robust to common artifacts. Third, data reliability: Using identifiability, we show MICAFlow maintains consistent performance across diverse datasets, including subjects with pathology, and is closely comparable to contemporary pipelines. In sum, MICAFlows combination of machine learning and efficient workflows produces research-grade data quality with clinical-grade speed. This work demonstrates that advanced MRI preprocessing can be done fast and robustly, helping close the gap between research neuroimaging and broad clinical application of quantitative MRI techniques. The source code for MICAFlow is available here: https://github.com/MICA-MNI/micaflow, and for LAMAReg here: https://github.com/MICA-MNI/LAMAReg.

bioinformatics↗

bioRxiv2026

Spatial Neuroimmune Crosstalk Driving Perineural Invasion in Head and Neck Squamous Cell Carcinoma

BackgroundPerineural invasion (PNI) is a clinically important feature of aggressive head and neck squamous cell carcinoma (HNSCC) and is associated with recurrence and poor survival. However, the spatial organization and molecular programs that characterize tumor-nerve interactions in HNSCC remain incompletely understood. MethodsSingle-cell-resolution Xenium spatial transcriptomic data from 10 HNSCC patients were analyzed to define nerve-associated regions, cell-type composition, tumor-nerve proximity, and perineural transcriptional programs. Complementary Visium spatial transcriptomic datasets were used to assess broader transcriptomic features of nerve-associated regions, including epithelial-mesenchymal transition and inferred ligand-receptor signaling. Clinical relevance was evaluated using bulk transcriptomic validation cohorts from TCGA-HNSC and GSE65858. ResultsSpatial mapping identified a distinct nerve-proximal microenvironment characterized by depletion of mature dendritic cells and altered immune and stromal composition, consistent with localized disruption of antigen-presenting support at the tumor-nerve interface. HPV- tumors exhibited closer tumor-nerve proximity and a higher exploratory PNI index compared with HPV+ tumors, suggesting subtype-specific differences in nerve-associated tumor behavior. Differential expression analysis of perineural versus distal tumor cells identified a spatially enriched gene program, including NFE2L2, MDM2, and PPARG, that demonstrated a nerve-proximal gradient most evident in HPV- disease. A composite three-gene score was associated with worse overall survival in HPV- patients in TCGA-HNSC and validated in GSE65858, while showing limited prognostic value in HPV+ disease. Visium analyses provided complementary evidence of EMT enrichment and inferred tumor-nerve signaling involving neural guidance and adhesion-associated pathways. ConclusionsThese findings support a model in which PNI in HNSCC reflects a spatially organized, transcriptionally distinct tumor-nerve microenvironment, particularly in HPV- disease. The NFE2L2/MDM2/PPARG signature may provide a candidate biomarker for risk stratification and nominates pathways for future mechanistic and therapeutic investigation.

cancer biology↗

bioRxiv2026

Precision mapping and molecular contextualization of surgical outcome epicenters in temporal lobe epilepsy

AO_SCPLOWBSTRACTC_SCPLOWTemporal lobe epilepsy is the most common drug-resistant epilepsy, with surgical resection offering the primary path to seizure freedom. Despite standardized approaches, a substantial proportion of patients experience seizure recurrence, and the neurobiological substrates underlying these divergent outcomes remain unclear. We applied an individualized normative modeling framework to multimodal preoperative MRI data in a group that subsequently underwent surgical resection, to characterize patient-specific structural deviations and identify disease epicenters. Patients who became seizure-free exhibited spatially coherent abnormalities localized to the hippocampus and ipsilateral association regions, anchored in agranular limbic territories and enriched for genes linked to calcium-dependent signaling. Non-seizure-free patients, on the other hand, showed a more heterogeneous and distributed pattern of deviations, consistent with a "temporal-plus" network organization, and broader neuromodulatory dysregulation. Crucially, overlap between resected tissue and network-defined epicenters was closely associated with seizure freedom, independent of total resection volume. These findings provide a multiscale framework for precision surgical planning, shifting the focus from standardized tissue removal to targeted disconnection of patient-specific pathological hubs.

neuroscience↗

bioRxiv2026

Selective Auditory Attention Decoding with a Two-Node Wireless EEG Sensor Network

Selective auditory attention decoding (sAAD) enables neuro-steered hearing devices by identifying the attended speaker in a multi-speaker environment from neural activity recorded with electroencephalography (EEG). Despite algorithmic progress, practical deployment remains constrained by a lack of wearable, unobtrusive, and fully wireless EEG acquisition solutions. Therefore, this work aims to evaluate whether reliable sAAD can be achieved under realistic hardware constraints imposed by using a wireless EEG sensor network (WESN) consisting of miniaturized, galvanically isolated EEG sensor nodes. Here, we use such a WESN consisting of two synchronized, compact around-ear EEG sensor nodes worn bilaterally. Each node provides four local EEG channels derived from five pre-gelled electrodes, including a local reference. Sample-wise wireless synchronization of data from both nodes enables joint processing as an eight-channel EEG. On a newly recorded dataset acquired with this setup, correlation-based stimulus decoding achieves an average sAAD accuracy of 69.24% on 60s decision windows, comparable to wired around-ear EEG systems that measure long-distance scalp potentials. Hidden Markov model-based post-processing further improves to a steady-state accuracy of 77.17% with an average simulated attention switch detection time of 32.79s. Combining sensor nodes at both ears outperforms single-ear configurations, primarily by providing redundancy that increases robustness rather than by exploiting complementary spatial information. Finally, we show that a fixed bipolar configuration using four electrodes per ear, yielding three channels, suffices to maintain performance. These results demonstrate the practical feasibility of sAAD using a fully wireless, galvanically isolated around-ear WESN and establish a realistic performance benchmark under practical hardware constraints.

neuroscience↗

bioRxiv2025

Rapid and accurate quantification of viable Listeria monocytogenes with clonal specificity using microfluidic droplet digital PCR based technology

To overcome the technical bottlenecks in the precise quantification and molecular typing of viable foodborne pathogens, this study establishes a microfluidic droplet digital PCR (ddPCR) based method for rapid and accurate detection and quantification of viable Listeria monocytogenes with clonal specificity. In contrast to the time-consuming plate culture methods and unspecific rapid detection methods, the method in this study employs clonal complex (CC) specific primers and probe for strain-specificity and integrate the nucleic acid dye propidium monoazide (PMA) to effectively distinguish viable from dead bacteria. The rapid and precise quantification of viable bacteria is achieved through microdroplet counting. This method does not require DNA extraction, and the entire detection process takes only about 3 hours, with a quantitative detection limit of 3.3x102 CFU/mL, providing strong technical support for the risk monitoring of highly pathogenic specific types of L. monocytogenes.

microbiology↗

bioRxiv2025

Panax notoginseng saponins protect the blood brain barrier against oxidative stress by activating the Akap12/PI3K/AKT/AQP1 signaling axis

Oxidative stress is a key pathological factor that contributes to blood-brain barrier (BBB) disruption during ischemic injury. Panax notoginseng saponins (PNS), the primary bioactive components of Panax notoginseng, possess potent antioxidative and vasoprotective activities. This study aimed to elucidate the molecular mechanisms by which PNS regulate endothelial integrity and BBB permeability under oxidative stress conditions. Mouse brain microvascular endothelial cells (bEnd.3) were treated with hydrogen peroxide to induce oxidative damage, followed by intervention with PNS. Transcriptomic and proteomic analyses revealed that Akap12 (A-kinase anchoring protein 12) and Aquaporin 1 (AQP1) were key molecules involved in PNS-mediated endothelial protection. H2O2 exposure markedly suppressed Akap12 expression and inhibited the PI3K/AKT signaling pathway, leading to elevated AQP1 expression and increased BBB permeability. PNS treatment reversed these effects by restoring Akap12 expression, reactivating PI3K/AKT signaling, and reducing AQP1 levels. Functional validation further showed that AQP1 dysregulation altered the expression balance of tight junction proteins, including Claudin-5 and ZO-1, thereby impairing barrier function. Collectively, these findings demonstrate that PNS protect endothelial cells from oxidative injury by activating Akap12-dependent PI3K/AKT signaling and modulating AQP1-associated tight junction remodeling. This study provides new mechanistic insights into the molecular basis of PNS-mediated BBB protection and suggests potential therapeutic targets for oxidative or ischemic cerebrovascular diseases.

pharmacology and toxicology↗

bioRxiv2025

Diagnosing linearity along the carbon cascade in terrestrial biosphere models

Elevated carbon dioxide (eCO2) fertilises photosynthesis, driving an increase in terrestrial gross primary production (GPP). However, it is unclear how effectively increased GPP propagates along the "carbon (C) cascade" to increase net primary production (NPP) and vegetation C stocks (Cveg) in different plant compartments. Vegetation models simulate divergent C cycle projections and have been criticised for being overly photosynthesis-driven (source-driven), neglecting processes that lead to non-linear behaviour in response to the GPP increase, which may attenuate (or amplify) changes in NPP and vegetation C stocks. Here, we introduce an analytical framework to diagnose linearity (L) of the land C cycle as the ratio of relative changes in linked fluxes and pools and apply it to outputs from 16 models of the TRENDY v11 ensemble. We found widely varying patterns in L across models and for the different links. Six models showed a clear dominance of larger relative changes in NPP than in GPP in global simulations (LNPP:GPP >1 for >60% of gridcells), indicating increased carbon use efficiency under eCO2. Only three models had LNPP:GPP < 1 for >60% of gridcells. Four models showed a clear dominance of larger relative changes in steady-state Cveg than in NPP, while five models showed an opposite pattern - in both cases with a large spread of LCveg*:NPP across gridcells within models. Three models showed a larger relative increase in root C than in Cveg, while two models showed a clear dominance of the opposite pattern. Widely differing distributions of L across models and links reveal a strong influence of alternative process representations (nonlinear behaviour) in individual models. However, for all links, L deviations from 1 were roughly balanced across the model ensemble, leading to an overall linear behaviour of terrestrial C cycle representations.

ecology↗

bioRxiv2025

An Unexpected Role of Molecular Chaperones in Regulating the Excited States and Activities of Folded Enzymes

Molecular chaperones are recognized for assisting protein folding. Emerging evidence suggests that chaperones also interact with natively folded proteins. Yet their functional impact on folded proteins remains unclear. Here, we show that chaperones can directly modulate the conformational dynamics and catalytic efficiency of enzymes in their native states. Spy and Hsp70 increased lysozyme activity by altering conformational exchanges, and this effect extended across other chaperones and enzymes. Hsp70, Spy, Hsp104, and Hsp20 enhanced the catalytic activities of multiple folded enzymes, including alkaline phosphatase, DNA polymerase Pfu, endonucleases Cas12a/Cas13a, and xylanase. These findings uncover an unanticipated function of chaperones in regulating enzyme function expanding their mechanistic scope beyond folding assistance and suggests opportunities in enzyme engineering, diagnostics, and cellular regulation.

molecular biology↗

bioRxiv2025

ContrastQA: A Label-guided Graph Contrastive Learning-based approach for protein complex structure quality assessment

Despite recent progress, the Estimation of Model Accuracy (EMA) for protein complexes remains less advanced compared to that for protein monomers. A key challenge lies in effectively integrating both interface-specific and global structural information to accurately assess the quality of protein complexes. Here, we introduce ContrastQA, the first EMA framework for protein complexes that incorporates the proposed label-guided graph contrastive learning based on interface quality. By integrating a geometric graph neural network to model global structural features, ContrastQA effectively captures both local (interface-level) and global (structure-level) information for accurate model quality estimation. ContrastQA achieved ranking losses of 0.123 and 0.116 on the TMscore and GDT-TS metrics on the CASP16 dataset, which are 0.015 (10.9%) and 0.012 (8.7%) lower than the second-best EMA method with ranking losses of 0.138 and 0.128. Furthermore, for the more challenging CASP16 hard targets, our method achieved a TMscore ranking loss of 0.131, which is 0.023 (14.9%) lower than that of the second-best EMA method with a ranking loss of 0.154. Our study demonstrates the strong effectiveness of the label-guided graph contrastive learning module. These findings suggest that our graph contrastive learning framework serves as a valuable pre-training strategy for learning protein structure representations. The source code is available at https://github.com/Cao-Labs/ContrastQA. Significance StatementHigh-quality proteins generated by high-precision structural prediction methods for protein complexes helping biologists prioritize models for drug discovery and understanding disease mechanisms. Recent advances in predicting protein complex structures (like AlphaFold3) still struggle to match experimental precision. A critical bottleneck lies in accurately estimating the quality of predicted models-known as Estimation of Model Accuracy (EMA). Here, we introduce ContrastQA, implementing label-guided graph contrastive learning framework to explore the generalization of high-quality proteins. Our method demonstrates outstanding performance, provides an effective solution for assessing the structural quality of protein complexes.

bioinformatics↗

bioRxiv2025

brain network dynamics during multi-task demands predict children math achievement

Dynamic reconfiguration of neural network and flexible information integration across multiple tasks has been considered critical to characterize individual difference in complex cognition and general intelligence. A promising and underexplored question is how these neurocognitive processes related to childrens academic achievements, a hallmark of high-order cognitive abilities that integrate attention, memory and problem-solving. By using of a multitasking paradigm which bridging outside- in and inside-out approaches, we investigated the dynamic neural mechanisms underlying two core domains of academic performance: math and reading. We first apply partial least squares regression (PLSR) to examine static neural patterns and find that the first latent component--reflecting a generalized brain functional system--predicts math achievement but not reading. The multiple-demand system and the somato-cognitive action network (SCAN) are consistently engaged across diverse task demands. Furthermore, we use a Hidden Markov Model (HMM) to examine dynamic features of brain activity and identify distinct integrated and segregated brain states. Notably, the segregated state--characterized by heightened cortical network segregation--is associated with better math performance. Information-theoretic analyses further reveal that greater complexity in the temporal sequence of the segregated brain functional networks, along with stronger cerebrocerebellar functional coupling, correlates with higher math achievement. By means of multitasks design, these findings suggest that flexible engagement of specialized brain network and automatic information processing is crucial for math learning in children.

neuroscience↗

bioRxiv2025

Advancing EEG Biosensing: Novel Scalable Production of Platinum-Silicon Microneedles

This study presents the development and analysis of a silicon microneedle electrode array fabricated using a streamlined, two-step approach for enhanced bio-signal recording applications. Numerical simulations with COMSOL Multiphysics were employed to optimize the microneedle array design, focusing on mechanical stability and minimizing skin-electrode impedance. Structural analysis identified needle lengths between 500 m and 700 m as optimal for mechanical robustness, with critical load factors indicating a high resistance to buckling under applied forces. Optimal needle spacing was determined to prevent inter-needle interference and ensure effective skin penetration. Proof-of-concept electroencephalography testing confirmed that the microneedles achieved reliable bio-signal recording, comparable to traditional wet electrodes, while offering advantages such as reduced preparation time and improved user comfort. This work establishes a simple and cleanroom-compatible, scalable method for fabricating platinum-coated silicon microneedle arrays, advancing the development of wearable, user-friendly biopotential monitoring devices.

bioengineering↗

bioRxiv2025

Synchronized EEG with two galvanically-separated miniature wireless behind-the-ear EEG sensors

We present a wireless EEG sensor network consisting of two miniature, wireless, behind-the-ear sensor nodes with a size of 2 cm x 3 cm, each containing a 4-channel EEG amplifier and a wireless radio. Each sensor operates independently, each having its own sampling clock, wireless radio, and local reference electrode, with full electrical isolation from the other. The absence of a wire between the two nodes enhances discreetness and flexibility in deployment, improves miniaturization potential, and reduces wire artifacts. A third identical node acts as a USB dongle, which receives and synchronizes the data from the two behind-the-ear nodes. The latter allows to process the 2 x 4 channel EEG as if all 8 channels are sampled synchronously, allowing the use of signal processing algorithms that exploit inter-channel correlations. To demonstrate this synchronized processing, we recorded auditory steady-state responses (ASSRs) at both ears and processed them with data-driven multi-channel filters to optimize the ASSR signal-to-noise ratio, demonstrating a more reliable ASSR detection compared to a single-ear setup.

bioengineering↗

bioRxiv2025

A wireless, scalable and modular EEG sensor network platform for unobtrusive brain recordings

This paper introduces a modular sensing platform for wearable electroencephalography (EEG) recordings. The platform is conceived as a wireless EEG sensor network (WESN), consisting of multiple miniaturized, wireless EEG sensor nodes that synchronously collect EEG data from different scalp locations. As there are no wires between the different sensors, the platform provides maximal flexibility and discreetness, combined with a reduced sensitivity to motion artefacts or electro-magnetic interference. By removing the driven right leg (DRL) electrode and reducing the within-node electrode spacing to 3cm, we obtain a compact design while maintaining a high signal integrity. The WESN system was validated through a series of experiments: achieving synchronization of EEG data transmission across multiple sensor nodes and the detection of actual neural responses in EEG experiments. These results demonstrate the effectiveness and robustness of the proposed WESN platform, establishing it as a promising research platform for scalable, flexible, and discreet multi-channel EEG monitoring in ambulatory settings.

bioengineering↗

bioRxiv2024

GOBoost: Leveraging Long-Tail Gene Ontology Terms for Accurate Protein Function Prediction

MotivationWith the advancement of deep learning, researchers have increasingly proposed computational methods based on deep learning techniques to predict protein function. However, many of these methods treat protein function prediction as a multi-label classification problem, often overlooking the long-tail distribution of functional labels (i.e., Gene Ontology Terms) in datasets. To address this issue, we propose the GOBoost method, which incorporates the proposed long-tail optimization ensemble strategy. Besides, GOBoost introduces the proposed global-local label graph module and multi-granularity focal loss function to enhance long-tail functional information, mitigate the long-tail phenomenon, and improve overall prediction accuracy. ResultsWe evaluate GOBoost and other state-of-the-art (SOTA) protein function prediction methods on the PDB and AF2 datasets. The GOBoost outperformed SOTA methods across all evaluation metrics on both datasets. Notably, in the AUPR evaluation on the PDB test set, GOBoost improved by 10.71%, 35.91%, and 22.71% compared to the SOTA HEAL method in the MF, BP, and CC functions. The experimental results demonstrate the necessity and superiority of designing models from the label long-tail distribution perspective. Availabilityhttps://github.com/Cao-Labs/GOBoost Contactcaora@plu.edu

bioinformatics↗

bioRxiv2024

Multimodal integration of single cell ATAC-seq data enables highly accurate delineation of clinically relevant tumor cell subpopulations

Accurately deciphering tumor cellular heterogeneity is crucial for developing effective treatments. While single-cell epigenomics assays offer powerful insights into studying tumor heterogeneity beyond the transcriptional level, their application remains a significant challenge. To address this, we introduce Multimodal-based Analysis of scATAC-Seq data (MAAS), a method designed to identify functional tumor cell subpopulations and reconstruct their lineages from single-cell sequencing assay for transposase-accessible chromatin (scATAC-seq). MAAS uniquely integrates multimodal information, encompassing chromatin accessibility, copy number variations (CNVs), and single-nucleotide variants, through self-expressive multimodal matrix factorization. MAAS outperforms existing methods in accuracy and robustness on both simulated and real datasets. When applied to multiple glioma scATAC-seq datasets, MAAS uncovered a previously hidden tumor cell subpopulation resistant to temozolomide treatment. Furthermore, MAAS effectively detected clinically relevant subpopulations with low CNVs, such as those found in pediatric ependymoma and B-cell lymphoma. Additionally, we developed a novel MAAS-derived clinical signature that provides superior prognostic prediction than traditional clinicopathologic features across multiple cancer types. In summary, MAAS significantly advances the identification of clinically relevant tumor cell subpopulations, thereby accelerating the discovery of potential therapeutic targets.

bioinformatics↗