Biology subjects

Chen, J.

Publications and source records attributed to Chen, J..

At least 307 records · Page 17Linked to original sources

bioRxiv2025

DINOcell: Learning Generalizable Perturbation Effects through Self-Distillation

AO_SCPLOWBSTRACTC_SCPLOWPredicting cellular responses to therapeutics is a promising approach for novel target discovery. However, state-of-the-art computational models designed to predict perturbation effects struggle to generalize and outperform simple baselines. We present DINOcell, a weakly supervised framework that adapts self-distillation to single-cell transcriptomics for predicting perturbation effects. We demonstrate that DINOcell outperforms baselines in predicting the effects of single gene perturbations. Furthermore, DINOcell accurately predicts non-additive effects of combination perturbations, indicating its capacity to model complex genetic interactions. Finally, we show that DINOcell learns representations that capture biological signals and is a promising, generalizable approach for in silico perturbation modeling, providing a valuable tool for accelerating therapeutic target discovery.

bioinformatics↗

bioRxiv2025

Nuclear N-Glycosylation Redefines the Glycoscape and Directs Cell Identity

Glycans modify nucleic acids, proteins and lipids in fundamental biological processes, but have long been considered confined to the secretory pathway and cell surface. Here, we challenge this view with evidence that N-glycosylation, a ubiquitous glycan modification, occurs within cell nuclei across tissues and species, exhibiting cell type-dependent abundance and localization patterns. Using a multi-faceted approach, we show that numerous classical membrane and secreted glycoproteins localize to the nucleus with N-glycan modifications, exemplified by the cell adhesion molecule L1CAM. Mechanistically, we show that the N-glycosylated L1CAM is transported to the nucleus by KPNB1, where it transcriptionally modulates neuronal differentiation. Moreover, this nuclear N-glycosylation event contributes to neurodevelopmental disorders and is pharmacologically reversible. These findings revise the cellular geography of N-glycosylation and expand the known routes of protein trafficking, and highlight nuclear N-glycosylation pathways as potential therapeutic entry points for developmental disorders.

cell biology↗

bioRxiv2025

Cross-species etiologically informed stratification enables T cell receptor-based diagnosis of Parkinson's Disease

BackgroundDeveloping peripheral blood-based diagnostic models for idiopathic Parkinsons disease (iPD), particularly those leveraging the T-cell receptor (TCR) repertoire, has long been considered infeasible because patient-derived TCRs appear to lack convergent sequence motifs. We reasoned that this apparent absence of shared TCR features likely reflects both insufficient sample sizes and unaccounted immune heterogeneity within the iPD population. MethodsWe reconstructed TCR repertoires from the two largest iPD cohorts currently available, anchoring them with three mechanistically distinct mouse models to stratify patients into model-anchored informed subtypes. Within these data-driven subtypes, we trained subtype-specific, multimodal multi-instance classifiers. ResultsWe demonstrated that iPD patients can be successfully stratified into model-anchored informed subtypes. Furthermore, the trained subtype-specific, multimodal multi-instance classifiers achieved an Area Under Curve (AUC) exceeding 0.8 for both subtypes. ConclusionsOur findings underscore the critical role of disease stratification in enabling TCR repertoire-based modeling in neurodegenerative diseases.

neuroscience↗

bioRxiv2025

Prolonged Loss of Oxidative Phosphorylation and Mitochondrial Mass Characterize CD66b+ Leukocytes from Patients with Sepsis

IntroductionSepsis leads to expansion of myeloid-derived suppressor cells (MDSC) and their subtypes. These normally transitory MDSCs suppress T cell activation and alter T cell cytokine production while simultaneously promulgating systemic low-grade inflammation. Immune metabolism can shape cell responses, regulate immune suppression, and enhance effector activity. Although MDSC metabolism has been extensively studied in cancer, the metabolic phenotype of this heterogeneous population in sepsis remains unclear. Our goal was to assess metabolic flux in blood MDSCs during and after sepsis and to stratify these patients clinical features and outcome with differences in metabolic flux that may guide treatment decisions. MethodsPeripheral blood mononuclear cells (PBMC) from healthy subjects and sepsis patients at 4 days, 2-3 weeks, and 6 months underwent CD66b+ or CD3+ enrichment, followed by assessment of metabolic flux, flow cytometry, mRNA sequencing, and chromatin accessibility. ResultsMitochondrial basal oxygen consumption rates (OCR) and maximal oxygen consumption rates (SRC, spare respiratory capacity) were decreased in MDSC from septic patients at 4 days after infection and persisted for up to 6 months after sepsis onset. Sepsis was not associated with differences in glycolysis. In contrast, oxidative metabolism in CD3+ T cells was similar between sepsis patients and healthy subjects. Reduced MDSC oxidative metabolism was linked to adverse clinical outcomes. The decline in oxygen consumption from MDSCs in septic patients was also associated with significant reductions in MDSC mitochondrial content. Transcriptomic analysis of CD66b+ cells isolated from PBMC of healthy participants and patients with sepsis at 4 days, 2-3 weeks, and 6 months revealed 19 differentially expressed genes and three long non-coding RNAs as potentially responsible for this decline in mitochondrial mass. Specifically, NR4A3, NR4A2, and TAMLIN/NR4A1 expression, all critical for mitochondrial biogenesis, were persistently decreased with reduced chromatin accessibility indicative of gene silencing. DiscussionAfter sepsis, blood CD66b+ cells present with reduced mitochondrial mass and oxidative metabolism that continue at least 6 months after sepsis. These changes in mitochondrial function result from a reduced content of these organelles. We have also identified gene silencing, reduced gene expression of key transcription factors that regulate mitochondrial biogenesis, as well as increased long non-coding RNA as potential drivers of this unique metabolic phenotype. These results highlight the potential benefit of targeting metabolism in sepsis to promote immune homeostasis and recovery.

immunology↗

bioRxiv2025

Predicting and Controlling Collective Fate in Multicellular Systems

Collective behavior is a defining property of multicellular systems, where coordinated outcomes emerge from local cell-cell interactions. Yet the quantitative rules linking single-cell decision-making to tissue-scale organization remain poorly resolved. Here, we develop a quantitative framework that defines an order parameter predicting when initially disordered colonies undergo a transition to ordered fate alignment and when minimal, localized inputs can redirect their collective state. This analysis reveals a distinct control regime in which multicellular assemblies become susceptible to a single engineered "guide" cell. We validate these predictions by introducing guide cells that integrate into unperturbed colonies and redirect fate patterns within the theoretically defined control windows. Together, these results connect single-cell decision rules to emergent tissue-level organization and establish a generalizable biological control strategy in which a minority engineered subset can reliably redirect the developmental trajectory of a much larger multicellular population. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=65 SRC="FIGDIR/small/693804v1_ufig1.gif" ALT="Figure 1"> View larger version (24K): org.highwire.dtl.DTLVardef@37baeforg.highwire.dtl.DTLVardef@d1d86aorg.highwire.dtl.DTLVardef@d7375eorg.highwire.dtl.DTLVardef@1741e9a_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOSummary figure:C_FLOATNO Emergence, Scaling, and Control of Multicellular Collective Order. (A) Cell-number-dependent collective order. Sparse colonies lack coordination and yield disordered fate distributions; dense colonies exhibit coordinated, ordered outcomes. (B) Collective order rises with colony compactness ({propto} cell density) and collapses across colony radii R onto a single curve. (C) Once coordination emerges, a single "guide cell" can steer fate-analogous to a sheepdog guiding a flock. The herding efficacy exhibits a biphasic window versus cell number. C_FIG

systems biology↗

bioRxiv2025

A Population-Scale Single-Cell Atlas of the Human Heart Reveals Cellular Remodeling and Cell-Cell Communication in Aging and Cardiac Disease

Previous single-cell and single-nucleus heart atlases, often limited by small sample sizes, lack the statistical power needed for phenotype association analysis, particularly for cardiovascular diseases and cardiac aging. To address this, we integrated data from 436 samples across 12 single-cell studies, harmonized the corresponding sample metadata, and constructed a comprehensive heart atlas comprising 355,762 cells and 1,436,719 nuclei. Consensus annotation identified 10 broad cell types and 54 fine-grained subsets. Associating gene expression patterns and cell type proportions with phenotypic data, we identified NRG1-expressing endocardial cells linked to multiple cardiac diseases and found that interferon (IFN) response signatures mark aging in multiple heart cell types. Importantly, we also developed PopComm, a novel computational method for inferring ligand-receptor (LR) interactions from population-scale single-cell data and quantifying interaction strength for individual samples. Using PopComm, we revealed a close association between the IFN response state and altered cell-cell communication during cardiac aging.

bioinformatics↗

bioRxiv2025

Identification of Vector Organisms and Their Carried Pathogens by Metagenomic Sequencing

Vector surveillance is vital for public health, yet traditional methods often face limitations. This study assessed the use of metagenomic sequencing for analyzing mosquitoes, cockroaches, flies, and rodents from Zhongshan, China. The results revealed a high concordance between metagenomic and Sanger sequencing for species identification (Kappa = 0.667, P = 0.011). Metagenomic analysis identified vector-specific dominant bacteria, including Wolbachia in mosquitoes, Blattabacterium sp. in cockroaches, Bartonella tribocorum and Leptospira interrogans in rodents, and Vagococcus teuberi in flies. It also detected numerous classical pathogens, such as Enterococcus faecalis, L. interrogans, and Klebsiella pneumoniae. Among the identified fungi and viruses, 19 were human pathogens, while seven posed a threat to agriculture, notably Pepper mild mottle virus and Tomato brown rugose fruit virus. These findings systematically characterize the microbial communities associated with various vectors, highlight the potential of metagenomics for broad pathogen screening, and underscore its utility in large-scale vector surveillance and public health protection. ImportanceVectors are major carriers of numerous diseases and present risks fot cross-border transmission. Existing detection methods for vectors and their pathogens remain inefficient, insensitive, and prone to difficulties in species identification. Despite extensive efforts in vector identification, the potential threats posed by their associated pathogens remain unclear, demonstrating the need for more effective diagnostic strategies. Metagenomic sequencing offers a promising solution to this challenge. In this study, we applied metagenomic sequencing to vectors collected from urban and port areas in Zhongshan, China, to evaluate its efficacy in pathogen detection. Overall, our work contributes to a better understanding of the public health risks posed by vector-borne pathogens.

microbiology↗

bioRxiv2025

Single-cell heterogeneity in interferon induction potential is heritable and governed by variation in cell state

Type I and III interferons (IFNs) are among the first lines of defense against viral infection, yet they are generally only produced by a tiny fraction of infected cells. Here, we show that cellular heterogeneity in IFN induction potential upon treatment with immunostimulatory RNA is not due to variability in sensing of stimuli but instead is shaped by heterogeneity in tonic cell signaling state. Using complementary single-cell approaches, we found that baseline variation in the c-Jun N-terminal kinase (JNK) and activator protein (AP)-1 transcription factor families correlated with IFNL1 expression predisposition. We further show that drug-based inhibition of JNK signaling virtually eliminates the innate antiviral response to immunostimulatory RNA. Finally, we show that single cell heterogeneity in IFN induction potential is heritable and stably maintained over numerous generations. Together, our study emphasizes the influence of intrinsic variability in cell state on innate immune regulation and IFN induction heterogeneity.

systems biology↗

bioRxiv2025

A haplotype-resolved T2T genome assembly of Indigofera pseudotinctoria reveals the genetic basis of flavonoid biosynthesis in Chinese Indigo

Indigofera pseudotinctoria, commonly known as Chinese Indigo, is a multifunctional leguminous shrub widely distributed in East Asia, valued for its medicinal, ecological, and forage importance, and naturally adapted to drought and nutrient-poor soils. However, the absence of genomic resources has hindered insights into the genetic basis of its bioactive metabolites and adaptive traits. Here, we present a haplotype-resolved, telomere-to-telomere (T2T) reference genome of I. pseudotinctoria, which represents the first complete genome within the genus Indigofera. By integrating PacBio HiFi sequencing, Oxford Nanopore ultra-long reads, and Hi-C chromatin conformation sequencing technologies, we successfully constructed two gap-free haplotypes assemblies (647.38 Mb and 657.52 Mb), characterized by high BUSCO completeness and QV scores, and fully captured all telomeric sequences. Comparative analysis between haplotypes revealed extensive synteny but notable structural heterozygosity but notable structural heterozygosity. Approximately 46% of structural variations overlap with genic or regulatory regions, leading to allele-specific expression divergence. Between the two haplotypes, we identified 2,607 and 2,331 haplotype-specific genes, reflecting a complementary functional specialization in defense and repair versus metabolism and growth. Integrative transcriptomic and metabolomic profiling across six tissues reconstructed the flavonoid biosynthetic network and identified MYB, bHLH, NAC, WRKY, and ERF transcription factors regulating five key pharmacologically active flavonoids (Calycosin, Butein, Sulfuretin, Chrysoeriol, and Genistin). These results collectively uncover the haplotype-specific regulatory and structural basis of Indigofera pseudotinctoria and establish a high-quality genomic framework for evolutionary, medicinal, and metabolic engineering studies in leguminous plants, and provide a framework for molecular breeding and bioactive compound discovery in medicinal legumes.

genomics↗

bioRxiv2025

Hepatocyte and adipocyte CDO1-mediated intracellular cysteine catabolism differentially modulates diet-induced obesity and fatty liver in mice

Background and AimsPlasma cysteine has been positively linked to BMI in humans. Consistently, dietary cysteine restriction significantly decreases obesity in mice via mechanisms that are incompletely defined. Cysteine dioxygenase type 1 (CDO1) mediates the major cysteine catabolism pathway and is highly expressed in hepatocytes and adipocytes. The goal of this study is to determine the impact of this endogenous cysteine catabolism pathway on obesity and fatty liver disease. MethodsDiet-induced obesity and fatty liver disease were studied in hepatocyte-specific CDO1 knockout mice (L-CDO1-KO), hepatocyte-specific CDO1 transgenic mice (L-CDO1-Tg), adipocyte-specific CDO1 knockout mice (Ad-CDO1-KO), and adipocyte-specific CDO1 transgenic mice (Ad-CDO1-Tg). ResultsDeletion of hepatocyte CDO1 decreased cysteine conversion to cysteine sulfinic acid, and had modest impact on liver cysteine, GSH, or taurine abundance. When fed a Western diet (WD), L-CDO1-KO mice showed elevated liver injury markers and inflammatory infiltration independent of obesity or steatosis. When fed a fibrogenic high fat/cholesterol/fructose diet (HFCFr), L-CDO1-KO mice developed worsened liver fibrosis. In contrast, L-CDO1-Tg mice fed WD showed lower blood glucose, but similar degree of obesity and steatosis compared to WT mice. Deletion of CDO1 in white and brown adipocytes of Ad-CDO1-KO mice had no effect on WD-induced obesity. In contrast, overexpression of CDO1 in white and brown adipocytes attenuated WD-induced obesity, which resulted in reduced hepatic steatosis. ConclusionGenetic manipulation of intracellular cysteine catabolism in hepatocytes and adipocytes differentially modulates diet-induced obesity and fatty liver, which warrants future mechanistic investigation to better understand cellular cysteine sensing mechanisms and cysteine control of cell metabolism.

physiology↗

bioRxiv2025

In situ profiling of nanoscale displacements uncovers mechano-architectural predictors of osteoarthritis emergence

Mechanical and anatomical interplay between the distinct tissues of the knee joint is essential for maintaining functional integrity during healthy ageing and contributes to the mechanisms that drive osteoarthritis (OA). In this study, we investigate how age- and disease-associated alterations in joint anatomy influence load transmission and tissue-level strain distribution. Using full-field synchrotron X-ray computed tomography coupled with digital volume correlation, we hierarchically characterised in situ nanoscale strains generated in response to mechanical loading across the tibial epiphysis. Our findings show that greater compressive strains accumulate in the articular condyle of male OA-prone (STR/Ort) epiphyses. Finite element modelling further demonstrated that these strain concentrations are associated with reduced load-bearing capacity, which arise from architectural differences localised to the subchondral bone plate. By coupling high-resolution imaging with computational modelling, our work provides new insights into how structural-function changes to joint anatomy contribute to the initiation and progression of mechanically driven OA. Our approach offers a means to identify early imaging biomarkers prior to OA diagnosis and has potential for monitoring interventions aimed at preserving joint mechanics while promoting healthy joint ageing.

bioengineering↗

bioRxiv2025

CBL mutations in pediatric solid and CNS tumors are a marker of receptor tyrosine kinase activation and a potential therapeutic target

Mutations in CBL, an E3 ubiquitin ligase that negatively regulates receptor tyrosine kinases (RTKs) and potentiates intracellular signaling, have been extensively characterized in hematological malignancies. However, the impacts are unknown in other cancer types. We have identified established and novel aberrations in CBL in molecularly diverse pediatric CNS and solid tumors. Additionally, we present novel CBL splice mutations, including in the germline of a high-grade glioma patient, and alternative CBL splicing without corresponding genomic alterations. Functionally, we demonstrate that CBL exon 8/9{Delta}, typically associated with acute myeloid leukemia, is RTK activating in neuroblastoma and that novel CBL variants, CBL E366_E373del and CBL C384G, can cooperate with overexpression of EGFR to transform Ba/F3 cells. Collectively, these data represent a novel form of RTK activation in pediatric patients with solid and CNS tumors who might benefit from RTK-targeted therapies. Statement of significanceMolecularly targeted therapies improve childhood cancer outcomes. Therapies targeting genes called receptor tyrosine kinases are particularly effective. We have identified new genetic alterations that may be a marker for receptor tyrosine kinase activation and drug sensitivity, enabling more patients to benefit from targeted treatments.

cancer biology↗

bioRxiv2025

Trichome formation in Nicotiana benthamiana is induced by Agrobacterium

Plant trichomes are specialised epidermal appendages found on various organs of many plant species. They play important roles in plant protection against pests, pathogens, herbivores, resistance to abiotic stress, light perception, mechanosensitivity etc. They are of direct agricultural importance, providing fibre (e.g. cotton) and producing secondary metabolites essential for agriculture, medicine, and industry. Trichomes can be relatively sparsely distributed, making their biological investigation and commercial exploitation more challenging. Here we show that infiltration of Nicotiana benthamiana leaf tissue with certain nopaline-type Agrobacterium tumefacians strains induces extremely dense, localised formation of glandular trichomes 15 days post infiltration (dpi). This Agrobacterium-strain specific effect was shown to be due to the presence of a trans-zeatin synthase (tzs) gene located on some Ti plasmids, which results in the bacterial production of the cytokinin trans-zeatin. This simple procedure enables dramatically increased trichome density to be compared biologically with adjacent isogenic plant tissues on the same leaf.

plant biology↗

bioRxiv2025

Persistent CD4+ T cell functional deficits during recovery from prolonged symptomatic SARS-CoV-2 infection

Symptoms of acute SARS-CoV-2 infection often resolve quickly but are sometimes associated with persistent immune dysfunction. The factors that predispose individuals to compromised immune function have not been well defined. We investigated CD4+ T cell phenotype and function in a small cohort of individuals who recovered from mild to moderate SARS-CoV-2 infection without hospitalization and were divided into short or prolonged symptom duration groups. Five individuals with prolonged symptom duration showed marked downregulation of CD4 on CD3+CD8- T cells (CD4low group) and a poor response to TCR stimulation with the superantigen Staphylococcal enterotoxin B (SEB), as shown by weak upregulation of the activation markers CD134 and CD69. CD4 surface intensities recovered to normal levels in four of these individuals within 3-12 months. Selected cytokines (IL-1RA, IL-7, and VEGF) were elevated in individuals with low CD4, but plasma levels of anti-S1 IgG did not correlate with CD4 defects. Bulk RNA sequencing of unstimulated and SEB-treated CD3+CD8- T cells revealed a >50% reduction in the number of differentially expressed genes in the CD4low group compared to the same individuals after CD4 levels were recovered and a healthy control group. Upstream regulator analysis of differentially expressed genes in unstimulated CD4low cells suggested a response to IFN, while SEB-stimulated CD4low cells showed reduced functionality of IL-2, CD28, and SATB1 regulated pathways. In summary, prolonged symptomatic recovery from SARS-CoV-2 infection was associated with a global CD4+ T cell response defect, defined by low surface CD4 expression, evidence of IFN signaling, and defective T cell activation.

immunology↗

bioRxiv2025

AI-based Predictive Signaling Pathway Profiling in Cardiac Fibrosis Suggests a Novel Combinatorial Treatment Strategy

BackgroundCardiovascular disease (CVD) remains the leading cause of global mortality, with myocardial fibrosis characterized by excessive extracellular matrix (ECM) deposition representing a common endpoint associated with progressive cardiac dysfunction. While studies in animal models of heart disease suggest that preventing or reducing fibrosis can antagonize negative ventricular remodeling, current therapeutic strategies remain clinically limited and ineffective, in part due to an incomplete understanding of the multifactorial nature of the fibrotic process. Methods and ResultsWe employed artificial intelligence (AI) trained from 6,528 cardiac fibrosis publications over the past decade, which suggested 10 nodal signaling pathways underlying the fibrotic process. Single-cell RNA sequencing was used to quantify pathway activities across mouse models and clinical samples encompassing acute and chronic cardiac injury. Dynamic enrichment analysis revealed 2 critical pathways involved in acute myocardial infarction (MI) injury driven temporally-regulated fibrosis whereby Janus kinase - Signal transducer and activator of transcription (JAK-STAT) signaling peaked early (day 7) while transforming growth factor-{beta} (TGF-{beta}) pathway activity peaked mid-phase (day 14). With more chronically-driven fibrosis, JAK-STAT signaling again emerged, which this time was persistently active during chronic transverse aortic constriction (TAC) injury. Experimentally, single-cell analysis identified a pathogenic myofibroblast subpopulation characterized by high JAK-STAT signaling and ECM secretion capacity. Fibroblast-specific Jak1/2 gene-deleted mice significantly reduced TAC-induced fibrosis by blocking myofibroblast formation from this subpopulation, although these same mice failed to show attenuated fibrosis following MI injury, suggesting a pathway that would be ideal to therapeutically target for chronic fibrosis without affecting necessary acute scar formation. Indeed, based on the predictive AI driven algorithm, a temporal inhibitory strategy was generated for JAK-STAT and TGF-{beta} that permitted compensatory scar formation while more effectively preventing progressive fibrosis and worsened cardiac function versus either singular pathway or chronic antagonism. ConclusionHere we employed the wealth of past signaling pathway data underlying cardiac fibrosis to train an AI model, which suggested an optimized approach of inhibiting 2 nodal signaling pathways but with differential timing as a more effective therapeutic strategy that reduces pathologic cardiac fibrosis without negatively impacting compensatory fibrotic activity with acute MI injury. The therapeutic approach involves altering the timing of JAK-STAT signaling blockade with ruxolitinib (RUX) in combination with delayed and temporary TGF-{beta} signaling inhibition post MI injury with pirfenidone (PFD).

cell biology↗

bioRxiv2025

Full-length structure of CPR-containing self-sufficient cytochrome P450

Cytochrome P450 (P450s) are heme-thiolate monooxygenases that exploit electrons sourced from pyridine nucleotide to reduce the oxygen and heme iron to catalyze hydroxylation of inactivated C-H bond1-4. Self-sufficient P450s that contain the substrate-binding heme-domain and the electron-donating NADPH:cytochrome P450 reductase (CPR) domain in the same polypeptide chain are highly effective and have been engineered to catalyze various challenging reactions. The high efficacy attributes to the effective electron transfer rate, but how the electrons travel among the redox centers remains elusive owing to the lack of structural information of the full-length protein. Here, we report the structure of a homologue of the most extensively studied P450BM3 from Shimazuella soli (SoP450) resolved by single particle cryo-electron microscopy (cryo-EM) and X-ray crystallography. SoP450 primarily exists as a homodimer formed via the intertwined CPR-domains. The spatial alignment of the heme-domain that is linked via an extensive loop was also determined. Notably, a class of structure that lacks one heme-domain was identified from the cryo-EM analyses, indicating that the heme-domain is mobile. We suspected that the heme-domain could move to reach the CPR-domain for electron acquisition and built a model of putative catalytic state to reveal how the electrons are relayed from NADPH to heme. These results are of fundamental importance to understand the catalytic reaction of CPR-containing self-sufficient P450s, which shall provide critical information to the engineering and applications of these enzymes.

synthetic biology↗

bioRxiv2025

Can We Extract Physics-like Energies from Generative Protein Diffusion Models?

Diffusion models have emerged as the state-of-the-art method in generative artificial intelligence (AI) and have shown great success in image synthesis, video generation, molecular design, and protein structure prediction. For biophysical problems, such as protein folding and association, a fundamental question in diffusion-based methods is how their learned functions correspond to thermodynamics. In this paper, we study diffusion models through the lens of theoretical biophysics, analyzing their underlying formulation of potentials and exploring their applications in scoring protein interactions. We develop simple theories rooted in statistical physics that relate thermodynamic potentials to the negative log of the probability of observing a system in a particular state. We include dimensional analysis of diffusion model equations and a table mapping AI and physics jargon. We then test a diffusion models ability to capture learned energies as negative log-likelihood values, -log p0(x0), by integrating over the diffusion-generated path or a probability flow path. We test these integrals on a simple 1D Gaussian mixture diffusion model and a protein-docking diffusion model, DFMDock. In the 1D case, we find that integration over both diffusion and flow paths can accurately recover ground truth probabilities. When we extract the learned docking energies for cases where DFMDock succeeds, we observe energy funnels with the minimum energy near the experimental docked structure, like those we observe with Rosetta, an empirically tuned physics-based biomolecular modeling suite. The learned energy performs comparably or outperforms Rosetta interface energy in 9 out of 25 cases at ranking the correctness of docked poses. These data show that we can extract a relevant learned energy function from a diffusion model and compare it to physical energy functions.

biophysics↗

bioRxiv2025

High-resolution MRI Guided Whole Mouse Brain Cell Type Atlas using Deep Learning

Cell types represent groupings of cells defined by shared anatomical and functional properties. Traditional mouse brain cell atlases rely heavily on single-cell sequencing, which provides valuable molecular detail but lacks whole-brain, isotropic resolution. Diffusion MRI (dMRI) offers a complementary approach for probing cytoarchitecture and myeloarchitecture, with quantitative metrics increasingly used as biomarkers of brain development and neurodegenerative disorders. Although prior work has linked dMRI metrics with gene expression, the capacity of dMRI to directly predict cell types remains unclear. Here, we present a deep learning framework that integrates high-resolution dMRI with three-dimensional light-sheet microscopy (LSM) registered to the Allen Mouse Brain Common Coordinate Framework (CCFv3). We investigated correlations between dMRI and spatial transcriptomics-derived cell types and generated a whole-brain cell type atlas at 10 {micro}m isotropic resolution. Together, these results establish an efficient, high-resolution strategy for brain cell atlas generation and underscore the potential of advanced imaging to illuminate cellular mechanisms of the brain.

neuroscience↗