Biology subjects

Kellis, M.

Publications and source records attributed to Kellis, M..

At least 19 recordsLinked to original sources

bioRxiv2026

An integrated multi-omic atlas of the choroid plexus highlights conserved molecular programs of epithelial function

The choroid plexus (ChP) is a specialized epithelial tissue responsible for production of cerebrospinal fluid (CSF), maintenance of the blood-CSF barrier and regulation of CSF-borne signaling factors. However, despite the ChP's central role in CSF homeostasis, conserved molecular programs of ChP epithelial function across species and conditions remain incompletely defined. In this study, we integrated eight mouse and human single-cell/nucleus RNA-seq datasets to build a multi-species ChP atlas capturing major epithelial, stromal, vascular, and immune populations. Cross-dataset differential expression identified a conserved epithelial gene program encompassing ion transport, barrier function, ciliogenesis, and secretory signaling, while prioritizing understudied genes including CLIC6 and VAT1L. Co-expression network analysis resolved conserved epithelial meta-programs, and transcription factor regulatory inference identified OTX2, MSX1, and GLIS3 among conserved regulators of epithelial gene expression. We also generated a novel Visium HD spatial transcriptomic atlas of mouse fourth-ventricle ChP to orthogonally validate epithelial modules. Lastly, cross-tissue analyses revealed substantial transcriptional overlap between ChP and retinal pigment epithelium, including shared epithelial genes and OTX2-associated regulatory relationships. Together, these findings define conserved molecular programs of ChP epithelial specialization, nominate candidate regulators for functional investigation, and provide an integrated cross-species and spatial resource for studying ChP physiology.

neuroscience↗

bioRxiv2026

ProxiNet transfers spatially learned cellular proximity to dissociated single-cell transcriptomes

Spatial transcriptomics reveals cellular organization within intact tissues, whereas dissociated single-cell RNA sequencing provides broad transcriptomic coverage but loses information about cellular proximity and neighborhood structure. Here, we developed ProxiNet, a spatially supervised framework that learns transcriptomic signatures of pairwise cellular proximity from spatial reference datasets and transfers these relationships to dissociated single-cell transcriptomes. ProxiNet predicted cellular proximity across brain regions and spatial technologies, including zero-shot cross-technology transfer, and gradient-based attribution identified genes and broader transcriptional programs associated with proximity predictions. In spatial datasets with known coordinates, ProxiNet-derived cellular neighborhoods recovered reproducible tissue organization and anatomical structure, providing independent spatial validation of the inferred proximity relationships. Applying the spatially calibrated model to dissociated scRNA-seq revealed heterogeneous cellular neighborhoods with distinct cell-type compositions and candidate communication programs. In an Alzheimer dataset, ProxiNet further identified age-associated remodeling of inferred neighborhoods, including an AD-associated neighborhood at 8 months characterized by distinct astrocyte and neuronal transcriptional states and candidate intercellular communication programs. Together, these results establish pairwise cellular proximity as an interpretable and transferable representation for extending spatially learned tissue organization to dissociated single-cell transcriptomes.

bioinformatics↗

bioRxiv2026

GeneSIS: enhancing transferability of polygenic scores with variant-level gene-by-sex interaction effects

Advancing precision medicine requires accurate prediction of disease liability across populations and contexts. A major challenge is the limited transferability of polygenic scores (PGS) across genetic ancestry groups. We present GeneSIS (GENE and Sex Interaction Score), a supervised statistical learning framework for jointly modeling additive and context-dependent genetic effects at single-variant resolution directly from individual-level data. We analyze 406,659 individuals, including admixed individuals, in the UK Biobank and 1.3 million variants to develop predictive models for 99 complex traits. We report that ~8% of selected variables capture gene-by-sex (GxS) effects, validated by sex-stratified analyses. Modeling GxS effects improves prediction across 32 traits in non-European individuals. For predicting hip circumference in Africans, GeneSIS achieves a 3.7-fold improvement (p=8.0x10-7) over linear-only PGS and highlights biologically plausible hypotheses, such as pleiotropic GxS effects of GCKR (rs1260326) on anthropometry and menopause, as well as GxS pathway enrichments for interleukin-4 regulation. Overall, our results highlight the benefits of integrating context-dependent effects in human genetics studies.

genetics↗

bioRxiv2026

Sequence-to-function deep learning decodes human cis-regulatory evolution

Deciphering the regulatory consequences of sequence divergence across human evolution is essential to understanding the molecular basis of human-specific traits and disease. Although millions of derived alleles distinguish humans from great apes, only a small fraction are likely to influence human-specific traits. Previous studies have focused on regions of elevated sequence divergence, assuming that rapid evolution reflects functional adaptation, yet individual high-impact regulatory mutations evade such scans. Here, we apply sequence-to-function deep learning to predict chromatin accessibility across modern human, archaic hominin, and great ape personalized genomes, identifying lineage-specific cis-regulatory elements (linCREs) across diverse cellular contexts. Compared to conserved elements, linCREs are shorter, less pleiotropic, less conserved, and enriched in neurodevelopmental pathways. Many linCREs occur in regions with limited sequence divergence that acceleration-based approaches would overlook. We validate lineage-specific enhancer activity through luciferase reporter assays and demonstrate that a single motif-generating derived allele nominated by model interpretability tools drives a hominin-specific neurodevelopmental enhancer.

genomics↗

bioRxiv2026

HI-JEPA: A World Model of Molecular Organization Learned from Measured Proximity

Proteins act through the company they keep. Which molecules occupy the same nanoscale neighborhood in intact tissue determines what can physically interact, and disease rearranges those neighborhoods before it changes anything a sequence records. That quantity (measured proximity between molecular species in unperturbed tissue) has never been acquired broadly enough to train on. Published colocalization arrives study by study and never accumulates into a graph. The measurement has to be made rather than collected. We built ASCEND, a spatial computing platform that measures pairwise molecular proximity from expansion microscopy at molecular resolution in intact tissue, and applied it to 164 proteins across 37 imaged regions in five studies, spanning cultured neurons, isolated synapses and mouse cortex in disease and control. HI-JEPA is a representation trained on those measurements. Each protein is one embedding, trained to predict the embeddings of its measured neighbors in latent space; it never reconstructs its input and generates no negatives. A set of proteins measured in one neighborhood forms a configuration, which is the object the model perturbs and plans over. The representation performs operations a sequence model cannot. It names a protein from the bare geometry of a microscopy point cloud, matched against 234,048 deposited structures, at top-1 accuracy 0.748 against a chance rate of 1.0 x 10-5. It predicts physical interaction between sequence-dissimilar proteins that were both withheld from training at AUC 0.908, where ESM-C 6B reaches 0.514 against partner-count-matched negatives. It recovers a held-out complex member in the top 100 of 13,447 candidates at recall 0.954, against 0.514 for a ranking built from complex frequency alone. Asked which partners a knockout disrupts, it recovers the experimentally observed ones at recall@100 0.640; asked the same question about a different protein, with the ranking rule and denominators unchanged, it recovers 0.028, so the answer follows the action. Given 5xFAD mouse cortex with no disease label, no reward and no indication that amyloid is relevant, ranking 1,574 measured assemblies by their departure from wild type returns amyloid-{beta} bound to AMPA receptor subunits in nine of the top ten. Planning over the same configurations independently selects the same subunits (GluA2, GluA3, GluA4) and predicts that disrupting the PSD-95 scaffold worsens the configuration, both agreeing in sign with experiments the model never saw. Ablating the measured-proximity channel at training time degrades cross-scale partner recovery from median rank 14 to 68 while leaving navigation and within-scale dynamics intact; ablating the perturbation channel does the reverse. The cross-scale capability therefore comes from the measurement and not from having seen more data. The intended application is target nomination in diseases where sequence and structure supply no starting point. Note: This is a capability report. The architecture, the training procedure and the acquisition protocol are proprietary and are not described. Section 4.2 gives the evaluation protocol behind every number reported.

systems biology↗

bioRxiv2026

ProtJEPA: A Multimodal Joint-Embedding Predictive Architecture for Protein Biological World Modeling with Multi-TeacherModality-Attentive Fusion

Over 99.9% of known protein sequences lack experimentally validated functional annotations. We present ProtJEPA, a multimodal Joint-Embedding Predictive Architecture that trains a sequence-only student encoder to predict joint embeddings spanning ten biological modalities--sequence, structure, knowledge graph, protein interactions, literature, localization, tissue expression, GO function, anatomy, and disorder--requiring only sequence at inference. The key innovation is target whitening, which eliminates severe anisotropy in joint targets (mean cosine 0.984 to 0.086) and prevents representation collapse without covariance regularization. On 1,828 held-out dark proteins with zero primary Pfam family overlap with training, ProtJEPA achieves 58.07% Hit@10 on zero-shot GO retrieval (+2.80 pp, p = 0.020), 69.99% enzyme class accuracy (+9.64 pp, p < 0.001), and +11.87 pp subcellular localization at 1% labels (p < 0.001). Under realistic dark-protein deployment conditions where relational modalities are unavailable, ProtJEPA significantly outperforms naive concatenation of remaining modalities. Cross-domain evaluations on drug-target interaction and disorder prediction confirm transfer beyond training modalities, with the T1-only < ESMC < ProtJEPA ordering replicated across six independent tasks. Ablations establish that Phase 1 aggregator pretraining and target whitening are each independently load-bearing.

cell biology↗

bioRxiv2026

Deep interpretable learning of sample representations for characterizing disease states in single-cell transcriptomics

Single-cell transcriptomics technology offers unprecedented insights into molecular heterogeneity. However, capturing sample-level representations that reflect both systemic and cellular states remains challenging, especially when disease annotations are mostly available as coarse sample-level labels. Here, we introduce Phenoverse, an interpretable deep learning framework that learns sample-level disease state representations through cell type-aware residual encoding, prototype learning, and Perceiver-based aggregation. Applied to independent single-cell transcriptomic cohorts of COVID-19, Alzheimers disease, and systemic lupus erythematosus, totaling over 5 million cells, we demonstrate that learned sample representations enable disease state prediction and encode a continuous spectrum of disease severity on unseen data that correlate with multiple clinical and pathological measures, despite being trained solely on binary phenotype labels. Further, we demonstrate that trajectory-derived genes reveal cross-cohort molecular programs and show consistently higher reproducibility than traditional case-control comparisons. Finally, prototype learning provides intrinsic model interpretability and enables the characterization of cell type-specific disease states. Taken together, Phenoverse offers an interpretable disease-phenotyping approach to dissecting sample heterogeneity, and our results highlight its utility in translating complex single-cell transcriptomic data into patient-level biological insights.

bioinformatics↗

bioRxiv2026

Regulatory landscape of widespread stop codon readthrough in Drosophila

In stop codon readthrough (SCR), ribosomes continue to translate past the stop. While programmed SCR is rare in mammals, it appears remarkably widespread in insects, yet its regulation and biological significance remain poorly understood. Here, we systematically characterize SCR in Drosophila melanogaster by integrating ribosome profiling, comparative genomics, proteomics, and functional screening. We identify [~]1,400 SCR genes, reveal sequence features associated with programmed SCR, and distinguish constitutive and tissue-restricted modes. We identify conserved cis-elements sufficient to stimulate constitutive SCR and demonstrate that many remain functional in mammalian cells. Unexpectedly, the major distinction between insect and mammalian cells was not termination efficiency, but insect responsiveness to SCR-promoting regulatory elements. SCR was further associated with increased overall protein output, suggesting a function beyond proteome diversification. Our results establish SCR as a pervasive regulatory mechanism in Drosophila and provide a new framework for its study through the Drosophila Stop Codon Readthrough Atlas.

genomics↗

bioRxiv2026

Circadian misalignment underlies immune escape in breast cancer

Circadian regulation shapes tissue physiology, yet how it organizes cellular and molecular dynamics within the tumor microenvironment (TME) and influences tumor immunity remains unclear. Using temporal single-nucleus multiomic profiling of mouse breast tumors, we uncovered extensive circadian programs that are both cell-type-specific and shared across the TME, governing proliferation and immune responses. Notably, cancer epithelial cells exhibited global acrophase misalignment relative to immune populations. This intercellular desynchrony manifests as temporal decoupling between tumor proliferation and immune activation, discordant antigen presentation and T cell recognition with intrinsic activation-exhaustion overlap in T cells, and asynchronous PD-1/PD-L1 oscillations that sustain checkpoint-mediated suppression. Similar patterns were observed in human triple-negative breast cancer (TNBC). Together, these findings establish intercellular circadian misalignment as a mechanism of tumor immune evasion and position the circadian architecture of the tumor-immune ecosystem as a previously underappreciated determinant of tumor development and therapeutic response. HIGHLIGHTSO_LISingle-cell multiomics maps circadian regulation of tumor-immune programs. C_LIO_LICircadian regulation in cancer epithelial cells is misaligned with immune cell populations. C_LIO_LITumor-immune temporal misalignment undermines antitumor immunity. C_LIO_LICircadian misalignment in human TNBC suggests relevance for immunotherapy timing. C_LI

cancer biology↗

bioRxiv2026

Length-dependent hypercontractility decoupled from cellular adaptations is the initial phenotype in MYBPC3-haploinsufficient iPSC-derived engineered cardiac tissues

Hypertrophic cardiomyopathy (HCM) is often characterized by a complex landscape of secondary compensatory changes, making it difficult to distinguish primary events triggered by the underlying genetic variants. Here, using patient-derived iPSC-engineered heart tissues cultured in conditions minimizing external stimuli, we identify the initial phenotype of the HCM inducing variant MYBPC3-Q1061X. We demonstrate that initially MYBPC3 haploinsufficiency leads to robust, muscle length-dependent hypercontractility with emerging diastolic dysfunction at higher contraction rates. This phenotype is entirely decoupled from early transcriptional and metabolic stress responses, as evidenced by physiological phenotype and metabolite and cell type-specific transcriptional profiles. Mathematical modeling reveals that the hypercontractile phenotype can be translated to adult cardiac muscle with a small change in myosin availability, amplified by inherent cooperativity within the sarcomere that is instrumental in enhancing the force and the speed of the contraction cycle. Despite the apparent calcium sensitization, these sarcomeric changes do not affect excitation-contraction coupling or calcium buffering in this proximal stage. These results suggest that the primary biophysical defect in MYBPC3-haploinsufficiency manifests as a latent mechanical hypersensitivity that precedes secondary cellular remodeling and functional instability. This initial phenotype provides a window for preventive therapeutic intervention before the onset of permanent cellular remodeling in the heart.

physiology↗

bioRxiv2026

Expansion Revealing of Pathology Resolves Nanostructures Associated with Inflammatory Phenotypes in COVID-19 Decedent Human Brain Tissue

Expansion revealing (ExR) elucidates cellular organization by separating proteins within dense nanostructures by 20x linear expansion, but requires fixation procedures incompatible with human pathology specimens. Here, we report ExR of pathology (ExRPath), which attains [~]20 nm resolution and decrowding of such tissues, through iterative 20x expansion, adapted to human brain pathology specimens. We also report a single-shot 15x expansion protocol for such tissues (15ExMPath), achieved through one-shot 15x expansion. Applying ExRPath and 15ExMPath to COVID-19-decedent brain tissue reveals periodic amyloid nanoclusters that co-localize with SARS-CoV-2 in a rare minority of patient specimens, pointing to a potential neuroinflammatory phenotype associated with COVID-19, and highlighting the power of high-throughput nanoimaging, empowered by expansion microscopy, for discovering potential novel disease mechanisms.

neuroscience↗

bioRxiv2026

Dissecting Alzheimer's disease heterogeneity by cross-trait polygenic prediction

Mapping the genetic basis of inter-individual heterogeneity in multifactorial diseases opens the door to mechanistic insights and opportunities for targeted intervention. In Alzheimers disease (AD), clinical and pathological heterogeneity is well recognized, but genetic dissection is limited by a lack of well-powered cohorts with deep phenotypic characterization. Here, we introduce a polygenic score (PGS) analysis strategy to address these limitations by leveraging the inherent pleiotropy in complex trait genetics. We perform a cross-cohort, cross-trait application of pre-trained PGS, integrating 713 UK Biobank-derived PGS with 36 deep AD phenotypes across 1678 ROSMAP participants. We identify 268 statistically significant (FDR<0.1) associations between 12 prioritized PGS and 36 AD phenotypes. Prioritized PGS include blood lipid measurements, inflammatory biomarkers, and cancer traits; observed AD phenotypes include cognition, amyloid, and tangles. Of the 268 associations, 49 persist with APOE-excluded PGS. Predictive models trained on multiple prioritized PGS outperform the AD PGS or APOE alone for predicting amyloid and cognition. Lastly, our approach identifies six individual-level AD polygenic subtypes supported by distinct pathological patterns. Overall, we combine large-scale biobank resources and deeply-phenotyped cohorts using PGS, reveal genetic features underlying AD heterogeneity, and provide a general model for stratifying heterogeneous disease-focused cohorts using genomics.

genetics↗

bioRxiv2026

Epigenetically constrained astrocyte states underlie prefrontal cortex vulnerability in Down syndrome associated Alzheimer disease

Down syndrome (DS), caused by trisomy 21, confers a near-universal risk for Alzheimers disease (AD), yet individuals exhibit marked variability in cognitive decline, suggesting the presence of cellular mechanisms that modulate vulnerability and resilience. However, these mechanisms remain poorly defined in the human brain. Here, we integrate matched single-nucleus RNA-seq and ATAC-seq profiles from the prefrontal cortex (PFC) and amygdala (AMY) of age-matched individuals with DS with and without AD (DSAD), enabling direct comparison within a shared genetic background. We identify basal astrocytes in the PFC as a selectively vulnerable cell state in DSAD, characterized by both reduced abundance and coordinated transcriptional and regulatory reprogramming. This state exhibits a shift away from homeostatic support functions, with decreased cytokine signaling and lipid-handling programs, alongside increased steroid- and nuclear receptor-associated activity. Concomitantly, chromatin accessibility profiling reveals reduced engagement of immune- and stress-responsive transcription factor programs, including AP-1, STAT, and BACH families, with linked regulatory perturbations at loci such as ABCA1, DAB2IP, and IL1RAP. Together, these findings define a previously unrecognized astrocyte state marked by epigenetic constraint and diminished responsiveness to stress and inflammatory signals, distinguishing it from classical reactive astrocyte phenotypes. Our results nominate PFC basal astrocytes as a key locus of vulnerability in DSAD and suggest that failure to mount appropriate astrocyte responses, rather than overt activation alone, may contribute to neurodegenerative progression.

bioinformatics↗

bioRxiv2026

Scn4b Modulates Huntington's Disease Phenotype Severity in vivo

Although it has been known for over 30 years that CAG trinucleotide repeat expansions in the HTT gene are the cause of Huntingtons disease (HD), it is still not understood how these mutations lead to the loss of striatal spiny projection neurons (SPNs) and other vulnerable neuronal cell types in HD. Here we show that SCN4B, a gene that is enriched in neurons that influence motor function, including striatal SPNs, modulates HD-associated phenotypes in vivo. Loss of Scn4b in wild-type mice mimics and Scn4b overexpression in an HD mouse model rescues several HD-associated phenotypes, including motor and cognitive deficits. Single nucleus RNA sequencing (snRNA-seq) analysis reveals that loss of Scn4b replicates several HD-associated gene expression signatures in the striatum. Conversely, overexpression of Scn4b rescues HD gene expression signatures and improves various SPN electrophysiological properties in HD model mice. Taken together, our results implicate loss of Scn4b expression as an important contributor to HD pathogenesis and therapeutic target in HD.

neuroscience↗

bioRxiv2026

Unravelling genome-wide mosaic microsatellite mutations at single-cell resolution

Short tandem repeats (STRs), or microsatellites, are highly mutable genomic elements that modulate gene regulations and are implicated in a range of human diseases. However, detecting mosaic STR mutations at single-cell resolution remains challenging due to both technical and biological complexities. To address this, we developed BayesMonSTR, a robust algorithm that enables accurate detection of mosaic STR mutations. Using this tool in single-cell analysis of human tissues, we reveal an accumulation of longer mosaic STR insertions and deletions (indels) in aging mitotic and post-mitotic cells. Strikingly, prefrontal cortex (PFC) neurons accumulate a higher burden of STR mutations than B cells or lung epithelium, with aged neurons exhibiting a particularly pronounced increase in longer STR deletions. These mutations are enriched at transcription start sites (TSSs) and active enhancers of highly expressed genes. Our work establishes a foundation for genome-wide, hypothesis-free discovery of disease-associated mosaic STR mutations and reveals a previously unexplored landscape of mosaic STR variation in development and aging.

bioinformatics↗

bioRxiv2026

Rectifying Stress into Structural Memory Drives Specialization in Cells and Adaptive Computational Systems

Functional specialization in continuous systems requires balancing adaptation to environmental stress with the preservation of encoded information. Yet, the physical constraints governing how living systems reconcile selective information retention with energetic structural reorganization remain unclear-a trade-off that may offer transferable design rules for adaptive computing. Here we present a unified framework that connects cellular specialization to adaptive machine learning through principles of nonequilibrium optimization. We find that asymmetric fluctuations in key parameters act as tunable drives that, when phase-aligned, are rectified into durable functional specialization. Biologically, we show that in adipose mesenchymal cells, transient nutrient deficits trigger cytosolic pH fluctuations. These oscillations, regulated by the ATP12A proton pump, modulate nuclear compartment dynamics to coordinate mitochondrial specialization and lock in epigenetic memory for phenotype persistence. Notably, modulating ATP12A amplifies these drives, enhancing functional specialization and metabolic homeostasis in disease models. Computationally, we formalize analogous principles using a compositional framework based on operadic algebra, where structural updates act as a topological ratchet converting drive fluctuations into persistent architectures. This reveals a learning regime where gains in predictive information are accompanied by auditable residual costs for non-predictive states and reorganization. As a proof of concept, we deploy a self-organizing machine learning system that discovers latent hierarchy and reconstructs data through adaptive compartment dynamics. We show that internal organization fluctuations increase with residuals, an information-processing ledger guiding adaptive computation. Across both domains, local organization gains are directionally linked to dissipation proxies (pH variance in vivo and residuals in silico), consistent with operation under shared constraints. Overall, this work identifies asymmetric-drive rectification as a fundamental mechanism of adaptive specialization, providing thermodynamically informed design rules for both bioengineering and adaptive AI. One-Sentence SummaryAsymmetric-drive rectification is a fundamental mechanism by which cells and adaptive computational systems convert internal stress into structural memory, encoding functional specialization

biophysics↗

bioRxiv2025

Convergent Signals Encode Persistent Memory of Resilient States

Caloric restriction reduces metabolic disease and associated comorbidities. Yet, the molecular mechanisms encoding cellular memory of these benefits remain unclear. Here, we use a functional genomics approach to integrate evolutionary cues, single-cell sequencing, and metabolomics, identifying convergent signals that encode epigenetic memory of adaptive states in visceral adipose tissue natural killer (NK) cells. Cross-species analysis shows that genomic hubs linked to human accelerated regions (HARs) are conserved and active during dietary restriction, forming transcriptional compartments that regulate oxidative stress response and DNA repair genes. Targeted screens and multimodal profiling across human data and mouse models reveal convergent signals needed for persistent adaptation: cooperative transcriptional regulators (NRF2, CIRBP, NR4A2) at key HAR-linked genomic hubs, innate immune mediators (IL15-IL2RB), and metabolic cofactors from linoleic acid oxidation. Activity of these convergent signals reduces DNA damage and methylation and enhances epigenetic plasticity and cytotoxic functions, thereby decreasing tissue fibrosis and senescence while maintaining both local and systemic metabolic plasticity. Mechanistically, linoleic acid metabolism supplies acetyl-CoA to sustain H3K27ac epigenetic marks that maintain coordinated repair and cytotoxic programs after programming. IL15-IL2RB signaling links tissue metabolic state to NK cell function for coherent adaptive responses. We show that engineering of NK cells with this multimodal programming leads to long-term preservation of therapeutic phenotypes in metabolic aging models, rescuing systemic dysfunction through damaged cells clearance and leptin sensitivity. Overall, these findings establish that convergent signals encode persistent cellular phenotypes linked to metabolic plasticity, suggesting design principles for engineering therapeutic memory of cell function. One-Sentence SummaryA multimodal approach reveals how dietary and evolutionary cues can be harnessed to engineer resilient phenotype memory.

immunology↗

bioRxiv2025

sc4D: spatio-temporal single-cell transcriptomics analysis through embedded optimal transport identifies joint glial response to Alzheimer's disease pathology

A precise understanding of disease-associated spatio-temporal transcriptional dynamics is critical for nominating therapeutic interventions that drive desired perturbation responses. In disease contexts, pathological processes unfold across diverse cellular states, spatial environments, and timescales; however, current computational approaches have a limited ability to jointly model these complex dynamics or infer cellular trajectories from in silico perturbation experiments. Here, we present sc4D, a biologically interpretable spatio-temporal (4D) analysis framework for single-cell transcriptomics in disease studies, integrating autoencoder embeddings with optimal transport. Applying sc4D to longitudinal spatial transcriptomics samples from Alzheimers disease mouse models, we report known disease biology and novel, testable mechanisms, including late-stage microglia-astrocyte syncytium near amyloid-{beta} plaques. In silico perturbations predict the restoration of protective, anti-inflammatory microglia through CAMTA1 activation or Donepezil, as validated by independent experimental findings. Overall, our work highlights the critical benefits of joint spatio-temporal modeling for elucidating disease mechanisms and predicting candidate interventions to improve cellular response.

bioinformatics↗