Biology subjects

Min, J.

Publications and source records attributed to Min, J..

At least 19 recordsLinked to original sources

bioRxiv2026

Stepwise reorganization of chromosome conformation and nuclear organization during stem cell differentiation

Nuclear organization is a fundamental feature of cell identity and cell fate determination. Compared with differentiated cells, pluripotent stem cells exhibit markedly distinct nuclear architecture at multiple levels, including chromosome folding, lamina association, histone modification landscapes and nuclear body organization. Although individual aspects of this reorganization have been characterized, how and in what order these features are remodeled as cells commit to a lineage remains poorly understood. Here, we temporally map nuclear reorganization across five stages of an in vitro differentiation system of human embryonic stem cells through definitive endoderm and hepatoblast intermediates into hepatocyte-like cells. By integrating Hi-C, RNA-seq, ATAC-seq, ChIP-seq, and CUT&RUN data, we establish a genome-wide framework linking structural, epigenetic, and transcriptional changes across differentiation. Combined with immunofluorescence imaging, chromosome painting, and liquid chromatin Hi-C (LC-Hi-C) we show that nuclear reorganization proceeds in a stepwise and temporally ordered manner through three major transitions. In the first transition, chromosomes condense into defined territories concurrent with transient anchoring of centromere-proximal regions to the nuclear periphery. In a second transition, active and inactive chromatin segregate, leading to stronger compartmentalization. This coincides with deposition and peripheralization of H3K9me2-marked chromatin and morphological changes in nuclear speckles, while chromatin conformation at speckle-associated regions stabilizes. In a third transition, after lineage-specific genes are activated, chromatin interactions globally stabilize, establishing a more stable nuclear architecture on top of these earlier large-scale structural rearrangements. Together, our results define a stepwise framework for nuclear reorganization during human embryonic stem cell differentiation and reveal that the transition from pluripotency to a differentiated state proceeds through coordinated, temporally ordered structural events.

genomics↗

bioRxiv2026

The function of human PIF1 in G quadruplex formation and replication stress response at ALT telomeres

Cancers maintain their telomeres through two telomere maintenance mechanisms: 85-90% of cancers rely on telomerase (TEL+), while 10-15% of cancers adopt the Alternative Lengthening of Telomeres (ALT) pathway. The Break-Induced Replication (BIR) pathway plays a critical role in maintaining telomere length in the ALT+ cells. In both yeast and human, PIF1, a 5' to 3' helicase, is required for the robust activity of BIR. However, the extent of human PIF1 (hPIF1) involvement in the ALT pathway remains unknown. Here we showed that hPIF1 can be recruited to damaged telomeres in ALT+ cells. In addition, we demonstrated that inhibition of hPIF1 induced DNA damage and G quadruplex (G4) accumulation at ALT telomeres, leading to a moderate reduction of the mean telomere length. Most interestingly, we demonstrated that inhibition of hPIF1 also attenuates checkpoint activation, BLM recruitment, single-stranded DNA (ssDNA) formation, DNA damage, and G4s at telomeres in the FANCM deficient ALT+ cells. Finally, we showed that inactivation of hPIF1 affects the viability of both ALT+ and TEL+ cancers, suggesting that hPIF1 is a potential drug target for cancer therapy.

molecular biology↗

bioRxiv2026

Reprogramming VHL with molecular glues enables selective degradation of caspase-2

Molecular glue degraders (MGDs) reprogram E3 ligases to eliminate neosubstrates, yet their application has largely been confined to CRBN. Here, we identify caspase-2 as a new neosubstrate for von Hippel-Lindau (VHL), expanding the scope of VHL-based MGDs. Guided by a focused VHL ligand library design, we employed TurboID-based proximity labeling to discover stereoisomeric compounds (dCASP2-1 and dCASP2-2) that selectively recruit caspase-2 to VHL and promote its ubiquitin-proteasome system-dependent degradation. Further structure-activity relationship (SAR) studies yielded dCASP2-3 and dCASP2-4, which enhanced degradation potency (by 622-fold relative to dCASP2-1) and abolished enantioselectivity. Mechanistic mapping localized the degrader-induced interface to a two-helix region of the caspase-2 CARD domain, with residues H33, P34, and D100 essential for VHL engagement. Degron-guided computational modeling of the VHL/MGD/caspase-2 ternary complex provided structural insight into neosubstrate recognition. Together, we report the development of VHL molecular glues that selectively and potently degrade caspase-2, offering chemical probes to interrogate its functions in apoptosis and stress responses, while broadening the substrate landscape of VHL-based MGDs.

biochemistry↗

bioRxiv2026

Dynamically actuated reconfigurable topographical surface enables active control of implant-associated infections

Implant-associated infections are driven by bacterial biofilm formation and remain difficult to eradicate using conventional antibiotic-based strategies. Here, we present a dynamically actuated reconfigurable topographical surface (DARTS) that integrates intrinsically bactericidal nanoscale surface topography with programmable mechanical actuation to achieve durable, antibiotic-free infection control. Using a scalable bottom-up nanofabrication strategy, we generate tunable wrinkled MXene topographies that exhibit contact-mediated bactericidal activity against both Gram-positive and Gram-negative bacteria without chemical leaching. Integration with a soft robotic actuator enables reversible modulation of surface geometry, which synergistically enhances bacterial removal and killing, resulting in near-complete disruption of mature biofilms. Dynamic actuation further sensitizes released bacteria to antibiotic treatment. In a mouse subcutaneous implant infection model, DARTS with actuation achieves sustained suppression of bacterial burden and markedly improves host tissue outcomes. Remote, noninvasive actuation using near-infrared laser stimulation further highlights the translational potential of this platform for implantable antibacterial applications. Significance StatementImplant infections are difficult to treat because bacteria form biofilms that protect them from antibiotics and the immune system. Current materials often rely on chemical release, which can lose effectiveness over time. Here, we present a new surface that both kills bacteria and removes them. The surface uses nanoscale features to physically damage bacterial cells, while dynamic motion clears attached bacteria and biofilms. This allows continuous, chemical-free control of infection. In a mouse implant model, the system greatly reduced bacterial burden and improved tissue healing. This work introduces a new way to control bacteria using dynamic surface design and could be applied not only to medical implants but also to environmental, food, textile, and marine systems.

bioengineering↗

bioRxiv2026

A general model for the evolution of thermal performance curves with application to real time-series data

Thermal performance curves (TPCs) are widely used to investigate the effect of changes in body temperature on an organisms performance. Despite empirical evidence that temperature-dependent performance is ubiquitous across taxa, the field lacks models for how thermal performance evolves under realistic timeseries, genetic architectures, and physiological constraints. We address this gap by integrating a mathematical model with individual-based quantitative population genetic simulations. Our model can predict the evolutionary trajectory and shape of TPCs for any given thermal regime. Our model reproduces core properties of TPC evolution from previous studies such as the emergence of generalists in variable environments, but also explains why organisms may evolve TPCs that do not match their historical body temperature range. We uncover novel dynamics of adaptive tracking, the most notable being multi-generation lags between temperature and TPC parameters that can lead to unexpected correlations between the two. Our model predicts empirically observed patterns of adaptive tracking of critical thermal minimum in the invasive pest Drosophila suzukii, including individual-level variability and multi-generation lags with changing temperature. Our results also highlight the limitations of models that ignore factors that influence TPC evolution and individual variability, such as autocorrelation in temperature timeseries, effective population size, evolution of additive genetic correlation in TPC parameters, genetic architecture, and physiological constraints. Our flexible simulation model can incorporate these factors and help generate empirically testable hypotheses of how species will evolve in response to global climate change.

evolutionary biology↗

bioRxiv2026

Distinct Neural Systems Supporting Afferent and Efferent Autonomic Activity

Numerous studies have reported brain correlates of autonomic activity. However, ambiguity exists regarding whether those correlates reflect receiving or sending out autonomic signals. Additionally, little is known about how emotion and aging interact with brain-autonomic coordination. Based on time-varying heart rate variability (HRV) and blood oxygenation level dependent (BOLD) data (N = 104 younger; N = 51 older adults), we found insular and cingulate activity was negatively correlated with HRV during emotion regulation. We further examined the afferent and efferent nature of HRV-BOLD correlations during rest (N = 102 younger; N = 51 older adults). Functionally afferent regions where decreases in HRV were associated with subsequently increased BOLD activity included the posterior insula, postcentral gyrus and frontal pole. Functionally efferent regions where increased activity was associated with subsequently increased HRV included the anterior insula and cingulate. Information appeared to flow in one direction as the afferent regions activity Granger-predicted the efferent regions activity. Together, our findings suggest a feedback loop where decreased HRV increases the afferent regions activity, which activates the efferent regions, leading to increased HRV. Aging seems to affect this system; the efferent and afferent regions scarcely overlapped for younger adults, but did overlap for older adults.

neuroscience↗

bioRxiv2026

Tumor extracellular vesicle RNA profiling predicts treatment response in pediatric diffuse midline glioma

Detection of reliable markers of therapy response and drug resistance remains a major unmet need in brain cancer, as serial tumor biopsy is often not feasible. This challenge is particularly acute in diffuse midline glioma (DMG), a fatal pediatric brain tumor for which new targeted therapies are entering clinical use, yet tools for real-time molecular analysis of tumor evolution during therapy remain lacking. Here, we demonstrate that plasma tumor-derived extracellular vesicle (EV) profiling provides a minimally invasive and complementary approach for diagnosis and longitudinal molecular monitoring in H3K27M-mutant DMG. Across patient-derived tumor models and clinical plasma samples, EV mRNA levels correlated strongly with parental tumor transcriptomes. EV H3K27M mRNA enabled discrimination of DMG from non-DMG controls, and exploratory EV response-associated mRNAs were linked to radiographic response and progression-free survival in patients receiving ONC201. To enable tumor-selective EV enrichment and multiplexed molecular profiling within a clinically practical workflow, we developed a new integrated platform supporting same-day EV analysis from small plasma volumes. This work represents the first proof-of-concept demonstration of the diagnostic and treatment response relevance of tumor-derived EV RNA in pediatric DMG and establishes a generalizable framework for minimally invasive longitudinal molecular monitoring in diseases where tissue access is inherently limited.

cancer biology↗

bioRxiv2026

Extra-lineage tissue programs define the transcription states of human pancreatic cancer

Cancers acquire alternate transcriptional states as they evolve, but the origins, timing and determinants of this plasticity are poorly understood in many tumours. We investigated the transcriptional states of pancreatic cancer by integrating [~]1000 tumour-enriched genomes and transcriptomes from 464 patients combined with scRNA-seq, multiome profiling, and spatial proteomics. Four epithelial states covering the spectrum of lineage plasticity were identified (Classical-1, Classical-2, Basal-1, Basal-2). Comparing these states to normal and pan-cancer human single cell atlases showed each state reflects distinct tissue programs found in other malignancies. Single cell analysis uncovered that the main transcription state of this disease (Classical-1) emerges before KRAS mutations. Spatial proteomics from patients and cancer-free donors showed that the Classical-1 program emerges during acinar-to-ductal metaplasia, and also unexpectedly, in normal ducts without disrupting their morphology. Overall, these findings link the extensive lineage plasticity potential of this organ to the origins of the transcriptional states.

cancer biology↗

bioRxiv2026

UniST: A Unified Computational Framework for 3D Spatial Transcriptomics Reconstruction

Spatial transcriptomics (ST) enables the measurement of gene expression in its native spatial context, yet most ST datasets are acquired as two-dimensional (2D) sections. Consequently, the underlying three-dimensional (3D) organization of tissues is only partially observed, and 3D ST data generated from serial sections are typically sparse and heterogeneous, with substantial tissue loss and missing measurements. These limitations pose major analytical challenges for reconstructing coherent 3D tissue architecture, rather than issues of experimental scalability alone. Here, we present UniST, a unified generative artificial intelligence (AI) framework designed to computationally reconstruct dense and continuous 3D ST landscapes from sparse serial sections, without altering the underlying experimental ST technologies. UniST integrates three complementary modules: kernel point convolution with cross-attention layers for point cloud upsampling, optical flow-based interpolation for continuous slice reconstruction, and a graph autoencoder with implicit neural representations for gene expression imputation. Together, these components densify sparse slices, resolve discontinuities, and map spatial coordinates to high-dimensional transcriptomics. Across multiple ST platforms and tissue contexts, UniST accurately restored structural continuity and biologically meaningful expression patterns. In a mouse embryo dataset, UniST reconstructed a dense 3D heart architecture from sparsely sampled slices. In two 3D human cancer tissues, UniST recovered critical spatial features, including tumor-immune boundaries and tertiary lymphoid structures, that were fragmented in the original data. By providing a generalizable computational solution that complements existing ST acquisition protocols, UniST facilitates cost-efficient and scalable reconstruction of 3D ST landscapes, enabling more faithful investigation of tissue organization and disease biology.

bioinformatics↗

bioRxiv2026

Identification of Dynamic Genetic Influences on DNA Methylation from Birth to Adulthood

Whether and how genetic regulation of DNA methylation (DNAm) change across the lifespan remains unclear. Here, we map age-dependent methylation quantitative trait loci (longitudinal mQTLs) using linear mixed models applied to repeated blood DNAm measures from birth, childhood, adolescence and adulthood in the Avon Longitudinal Study of Parents and Children. We identify 2,210 longitudinal mQTLs (2,393 SNP-CpG pairs; 7.3% trans) and observe consistent genotype-by-age effects in two independent cohorts of diverse ancestries (Pearsons r = 0.85 in the Generation R Study; r = 0.56 in the Drakenstein Child Health Study). Longitudinal mQTLs show increasing effects with age at half of loci and associations with multiple phenotypes. CpGs with longitudinal mQTLs are more heritable and enriched in regulatory elements and pathways related to multicellular organism development and cell adhesion. These results chart dynamic genetic influences on the human methylome and provide a novel perspective on epigenetic regulation.

genomics↗

bioRxiv2026

A New Sparse Bayesian Quantile Neural Network-based Approach and Its Application to Discover Physiological Sweet Spots in the Canadian Longitudinal Study on Aging

Identifying physiological sweet spots (optimal ranges for homeostasis) is essential for precision medicine. However, traditional statistical methods often rely on globally linear or locally jagged models that struggle to capture the smooth, non-linear nature of biological regulation in high-dimensional data. We present the Quantile Feature Selection Network (Q-FSNet), a neural network-based framework that integrates quantile regression, feature selection, and uncertainty estimation to identify biomarkers with sweet spots. Unlike traditional methods, Q-FSNet learns continuous response curves without requiring pre-specified number of change points. We further introduce Quantile Dirichlet Network (Q-DirichNet), a fully Bayesian extension that utilizes Dirichlet priors to automate feature shrinkage. Using data from the Canadian Longitudinal Study on Aging, we identified 25 metabolites with distinct homeostatic ranges for which biological age acceleration is minimized. The metabolites with sweet spots for biological aging include some derived from diet or produced by the gut microbiome; this highlights their potential for knowledge translation and public health impact. Our results, corroborated by existing literature, demonstrate that these sparse neural network-based methods offer a scalable and interpretable tool for discovering metabolic signatures of healthy aging vs. dysregulation in large-scale omics research.

bioinformatics↗

bioRxiv2026

SMARCAL1 is a candidate therapeutic target for ALT-positive tumors

A significant subset of tumors, including over 50% of osteosarcomas--an aggressive bone malignancy affecting children, adolescents, and young adults--relies on alternative lengthening of telomeres (ALT), a telomerase-independent, DNA repair-based mechanism for telomere elongation. The overall 5-year survival rate for osteosarcoma patients is [~]65%, underlying the need to develop novel targeted therapies. Through the Cancer Dependency Map, we identify SMARCAL1, a DNA translocase previously shown to remodel stalled replication forks, as a top selective dependency factor in telomerase-negative tumors. Using a panel of ALT-positive and ALT-negative cancer cell lines, as well as osteosarcoma patient-derived xenograft cells, we confirm that ALT-positive cells are uniquely sensitive to the loss of SMARCAL1, whose depletion exacerbates ALT-dependent phenotypes and telomeric DNA damage. Notably, we demonstrate that suppressing ALT abrogates their dependency on SMARCAL1. Mechanistically, we show that SMARCAL1 loss leads to telomeric ssDNA accumulation in ALT-positive cells, dependent in part on DNA repriming mediated by the DNA primase/polymerase PRIMPOL. Moreover, SMARCAL1s ssDNA annealing activity counteracts DNA unwinding by the BLM helicase, limiting telomeric ssDNA accumulation and DNA damage in ALT-positive cells. Importantly, SMARCAL1 depletion induces senescence in ALT-positive cancer cells, rendering them susceptible to treatment with senolytic agents. Together, these findings establish SMARCAL1 as a key regulator of ALT metabolism and highlight SMARCAL1 as a promising therapeutic target for ALT-positive tumors.

cancer biology↗

bioRxiv2025

Neural posterior estimation for population genetics

Simulation-based inference methods are increasingly being used in population genetics due to their flexibility and ability to be applied in settings where likelihood-based methods are intractable. Perhaps the best known such method is Approximate Bayesian Computation (ABC); however, its popularity is offset by its shortcomings which include computational expense and an unfortunate inability to efficiently fit models to high-dimensional summaries of the data. An alternative approach that solves these issues is supervised machine learning (ML); however, ML methods generally do not yield Bayesian uncertainty estimates of the quantities they predict. Here, we apply a recently introduced method, neural posterior estimation (NPE), that combines the best facets of ABC and supervised ML by training a neural network to estimate the posterior distribution of a population genetics model. We first compare neural posterior estimation with other inference methods for a variety of population genetic tasks, and show that neural posterior estimators yield posterior distributions with high accuracy and efficiency. We compare learned posterior distributions given raw genotypes and various summary statistics as input data. Additionally, we apply neural posterior estimation for demographic inference for simple and more complex models to highlight its application, including an analysis of demographic history in Drosophila melanogaster. Finally, we provide a user friendly workflow that enables others to perform neural posterior estimation on their own genetic data.

evolutionary 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

Parallel but distinct adaptive routes in the budding and fission yeasts after 10,000 generations of experimental evolution

Quantitative genetics approaches designed to understand the evolution of traits have helped improve our understanding of the genetic basis of adaptation. However, they often overlook crucial aspects of adaptation, including the long-term temporal evolutionary dynamics, the predictability of evolutionary outcomes, the influence of past evolution on future evolutionary trajectories (contingency), and the diversity of molecular mechanisms underlying adaptation. Experimental evolution has been a useful tool for answering these questions, but extracting fundamental principles and predictive features of evolutionary outcomes from these datasets remains challenging due to the large number of covariates and confounding effects, such as differences in experimental setup, species lifestyle, gene content, and evolution rate. Here, we sought to circumvent these challenges by comparing distant yeast species that share several evolutionary features but differ mainly in evolutionary history and genome architecture, i.e. Saccharomyces cerevisiae and Schizosaccharomyces pombe. Thus, we evolved 10 populations of the fission yeast for 10,000 generations in the same conditions as a pre-existing budding yeast dataset (i.e. high-sugar media and hypoxic conditions), allowing us to observe repeatable evolutionary outcomes within species but diverse molecular mechanisms and targets of adaptation across species. The most frequent adapting route in these conditions involved upregulating fermentation genes and downregulating the glycolysis gene pyk1, which has not previously been observed in S. cerevisiae evolved populations or in wild Kluyveromyces lactis, but similar evolutionary paths have been observed in Schizosaccharomyces japonicus and in clinically relevant populations, such as some cancer cells. This suggests that parallelism is pervasive in the tree of life and that mechanisms of adaptation can be shared among closely related or distant species. Despite similar gene content and identical environments, recurrent adaptation across S. pombe populations involved different genes than in S. cerevisiae and was mostly detectable at the transcriptomic level. This suggests that trans-regulatory effects may play an important role in adaptation on short evolutionary timescales and that differences in evolutionary outcomes between these species may be attributed to contingency.

evolutionary biology↗

bioRxiv2025

Epigenetic signature of heterogeneity in aging: Findings from the Canadian Longitudinal Study on Aging

Background: Human aging does not follow a single trajectory. Epigenetic changes offer insight into the heterogeneity in aging by reflecting the combined influence of genetic, environmental, and lifestyle factors on the timing and progression of age-related changes beyond what chronological age alone can explain. Recent studies in cancer and aging underscore the importance of methylation variability as a marker of biological dysregulation. Methods: We investigated the role of DNA methylation in aging heterogeneity by performing epigenome-wide differential methylation and variance association analyses in blood samples from 1,445 Canadians aged 45 to 85 from the Canadian Longitudinal Study on Aging. Results: We identified 448 differentially methylated regions and 488 differentially variable regions associated with health decline as measured by the health deficit accumulation Frailty Index, cognitive function, and physical function. These two classes of regions showed minimal overlap, with distinct gene coverage, suggesting that variability contributes a complementary signal to aging heterogeneity. Genes overlapped by differentially methylated regions were enriched for immune and inflammation-related pathways, whereas differentially variable regions highlighted additional localized, CpG-island-enriched signals shared across health domains, consistent with regionally structured rather than diffuse dysregulation. By integrating significant CpGs from both analyses, we constructed an epigenetic biomarker. The biomarker was associated with all-cause mortality and showed higher discrimination than biomarkers constructed from differential methylation or variability alone, with a similar pattern reproduced in the Baltimore Longitudinal Study of Aging. Conclusions: These findings suggest that DNA methylation variability may provide a complementary dimension of epigenetic aging and support further evaluation in larger cohorts with more mortality events.

genomics↗

bioRxiv2025

ATOMIC: A graph attention neural network for ATOpic dermatitis prediction on human gut MICrobiome

Atopic dermatitis (AD) is a chronic inflammatory skin disease driven by complex interactions among genetic, environmental, and microbial factors; however, its etiology remains unclear. Recent studies have reported the role of gut microbiota dysbiosis in AD pathogenesis, leading to increased interest in microbiome-targeted therapeutic strategies such as probiotics and fecal microbiota transplantation. Building on these findings, recent advances in computational modeling have introduced machine learning and deep learning-based approaches to capture the nonlinear relationships between gut microbiota and diseases. However, these models focus on diseases other than AD and often fail to capture complex microbial interactions or incorporate microbial genomic information, thereby offering limited interpretability. To address these limitations, we propose ATOMIC, an interpretable graph attention network-based model that incorporates microbial co-expression networks to predict AD. Microbial co-expression networks incorporate microbial genomic information as node features, thereby enhancing their ability to capture functionally relevant microbial patterns. To train and test our model, we collected and processed 99 gut microbiome samples from adult patients with AD and healthy controls at Kangwon National University Hospital (KNUH). As a result, ATOMIC achieved an AUROC of 0.810 and an AUPRC of 0.927 on the KNUH dataset. Furthermore, ATOMIC identified microbes potentially associated with AD prediction and proposed candidate microbial biomarkers that may inform future therapeutic strategies. To facilitate future research, we publicly released a gut microbial abundance dataset from KNUH. The source code and processed abundance data are available from ATOMIC GitHub repository at https://www.github.com/KU-MedAI/ATOMIC. Author summaryAtopic dermatitis (AD) is a chronic inflammatory skin disease affecting approximately 120 million people worldwide, often leading to a reduced quality of life, including sleep disturbances and stress. Current treatment methods rarely achieve complete remission, and the precise cause of AD remains unclear. The association between imbalances in the gut microbiota and the pathogenesis of AD has highlighted the gut microbiota as a promising target for therapeutic strategies. Exploring the potential role of the gut microbiota in modulating host immune responses is likely to have a positive impact on future AD treatment. Consequently, the growing need to develop machine learning and deep learning-based predictive models using gut microbiome data. However, existing models often suffer from limited interpretability owing to their black-box nature and frequently overlook microbial interactions and genomic contexts. To address these limitations, we propose a novel model called ATOMIC, which integrates microbial co-expression networks and genomic information to predict AD. Our model achieved superior performance in terms of AUROC, AUPRC, and F1 scores compared with existing models and provided interpretable insights through an attention mechanism, thereby offering promising potential for identifying relevant biomarkers. Together, these results highlight ATOMICs potential for the accurate prediction of AD and for identifying biologically meaningful microbial signatures.

systems biology↗

bioRxiv2025

AI-powered Deep Visual Proteomics reveals critical molecular transitions in pancreatic cancer precursors

Pancreatic ductal adenocarcinoma (PDAC) evolves through non-invasive precursor lesions, yet its earliest molecular events remain unclear. We established the first spatially resolved proteomic atlas of these lesions using Deep Visual Proteomics (DVP). AI-driven computational pathology classified normal ducts, acinar-ductal metaplasia (ADM), and pancreatic intraepithelial neoplasia (PanIN) from cancer-free organ donors (incidental, "iPanINs") and PDAC patients (cancer-associated, "cPanINs"). Laser microdissection of 96 discrete regions containing as few as 100 phenotypically matched cells and ultrasensitive mass spectrometry quantified a total of 8,512 proteins from formalin-fixed tissues. Distinct molecular signatures stratifying cPanINs from iPanINs, and remarkably, many cancer-associated proteins already marked histologically normal epithelium. Four core programs - stress adaptation, immune engagement, metabolic reprogramming, mitochondrial dysfunction - emerged early and intensified during progression. By integrating DVP with AI-guided tissue annotation, we demonstrate that molecular reprogramming precedes histological transformation, creating opportunities for earlier detection and interception of a near-uniformly lethal cancer. SignificanceOur spatially-resolved proteomics atlas uncovers distinct molecular signatures in pancreatic cancer adjacent precursor lesions, clearly diverging from those in incidental, cancer-free pancreatic lesions. Our deep proteomics dataset offers a valuable resource for identifying novel biomarkers and therapeutic targets, informed by the earliest cancer-associated molecular events in archival pancreatic tissues.

cancer biology↗