Unsupervised Clustering Research Articles

Diverse stromal phenotypes at spatial resolution in colorectal cancer peritoneal metastasis.

218 Background: Peritoneal metastases (PM) in colorectal cancer (CRC) portend a poor prognosis. We sought to elucidate spatially resolved molecular features facilitating transcoelomic dissemination and progression. Methods: 50 PT, 116 PM and 37 primary normal samples from 96 patients were retrieved and profiled with digital spatial profiling (GeoMx DSP, Nanostring Technologies). Through unsupervised clustering, we characterized the microenvironment of tumor-stroma compartments and studied the roles of stromal phenotypes in promulgating tumorigenesis. These findings were orthogonally validated through spatial proteomics (multiplex IHC [mIHC] - COMET, Lunaphore Technologies) of 10 paired PT-PM samples retrieved from 6 patients. Results: A median of 37 region of interests (ROIs) (IQR: 32-38) were selected from a total of 9 tissue microarray (TMA) slides. A total of 269 ROIs (181 tumor, 88 stroma) from PM samples, 45 ROIs (34 tumor, 11 stroma) from PT samples and 28 (all stroma) ROIs from primary normal samples were retrieved and profiled. Through consensus clustering of stroma ROIs, we identify a fibro-collagenous and immune infiltrated stromal phenotype (stromal cluster [SC] 2) characterized by increased cancer associated fibroblasts (CAFs), memory B cells, M2 macrophages and T-cell exhaustion. Patients with SC2 stroma had poorer survival (p=0.036). SC2 stroma was also observed to support adjacent tumor compartments with enriched oncogenic pathways such as TGF-beta, TNF-alpha, hypoxia and JAK-STAT. Through discriminatory gene expression profiles, we developed a 20-gene composite bidirectional signature of SC2. The prognostic significance of the SC2-signature was externally validated through several cohorts including TCGA and MSI-H only cohorts. Next, by inspecting mIHC retrieved immune cell type densities across ROIs, we confirmed a similar phenomenon in which two distinct stromal clusters were found – one being fibroblast infiltrated but T-cell depleted (SC2-like, n=67 ROI) and another which is T-cell infiltrated but fibroblast depleted (SC1-like, n=104 ROI). Diverging spatial distributions of T-cells and fibroblasts between SC1 and SC2-like stroma was appreciated as well. Between both PT and PM, we find close clustering of T cells in SC1-like stromal compartments and fibroblasts in SC2-like stromal compartments. Conclusions: We describe SC2, a pro-tumorigenic stromal phenotype characterized by increased CAFs, T cell exhaustion and is associated with poor prognosis in CRC PM.

Top-view characterization of broiler walking ability and leg health using computer vision.

Impaired walking ability and leg health are commonly seen in broilers and can negatively impact their welfare. Commonly, walking ability and leg health are assessed manually, but this is time consuming and can be subjective. Automated approaches for scoring walking ability and leg health at the individual level could therefore have great added value. Here, we studied whether automatically extracted top-view walking features of broilers can be used as a proxy for walking ability and leg health. Top-view videos were collected of unmarked broilers walking through a corridor that was placed inside the home pen. From these videos, four top-view features were derived: 1) lateral body oscillation, calculated as deviations from the movement trajectory of the bird, 2) step count, 3) completion time, and 4) length-width ratio of the virtual bounding box encapsulating the bird while walking as an indicator of wing support. We assessed the relationship of these computer vision-based features with manual gait and leg health scores, including hock burn (HB) and footpad dermatitis (FPD). We observed that birds with worse gait scores (GS) had longer completion times, higher step counts and a trend for higher lateral body oscillation levels in the walkway setup. Unsupervised clustering using the K-means algorithm with these walking features showed potential to distinguish birds with GS3+, although differentiating between GS1 and GS2 proved more challenging. We concluded that the length-width ratio of the bounding box during walking was not a suitable proxy for poor gait. We found no relationship between top-view walking features and mild cases of HB and FPD in broilers. Overall, the results of this study indicate that top-view video recordings can provide insight into birds' walking ability, using features related to movement speed, step count and lateral body oscillation, making automated scoring more feasible on a larger scale in practice. However, these top view features provide little information about mild HB and FPD.

What makes us feel good? A data-driven investigation of positive emotion experience.

What does it mean to feel good? Is our experience of gazing in awe at a majestic mountain fundamentally different than erupting with triumph when our favorite team wins the championship? Here, we use a semantic space approach to test which positive emotional experiences are distinct from each other based on in-depth personal narratives of experiences involving 22 positive emotions (n = 165; 3,592 emotional events). A bottom-up computational analysis was applied to the transcribed text, with unsupervised clustering employed to maximize internal granular consistency (i.e., the clusters being maximally different and maximally internally homogeneous). The analysis yielded four emotions that map onto distinct clusters of subjective experiences: amusement, interest, lust, and tenderness. The application of the semantic space approach to in-depth personal accounts yields a nuanced understanding of positive emotional experiences. Moreover, this analytical method allows for the bottom-up development of emotion taxonomies, showcasing its potential for broader applications in the study of subjective experiences. (PsycInfo Database Record (c) 2025 APA, all rights reserved).

Discovering patient groups in sequential electronic healthcare data using unsupervised representation learning

IntroductionUnsupervised feature learning methods inspired by natural language processing (NLP) models are capable of constructing patient-specific features from longitudinal Electronic Health Records (EHR).DesignWe applied document embedding algorithms to real-world paediatric intensive care (PICU) EHR data to extract patient-specific features from 1853 patients’ PICU journeys using 647 unique lab tests and medication events. We evaluated the clinical utility of the patient features via a K-means clustering analysis.ResultsWe trained a document embedding model under a unique evaluation pipeline and obtained latent patient feature vectors for all 1853 patients. We performed unsupervised clustering to the patient vectors as a downstream analysis and obtained 5 distinct clusters via hyperparameter optimisation. Significant variations (p<0.0001) within both patient characteristics and surgery intervention and diagnostic profiles were detected.ConclusionThe K-means clustering results demonstrated the clinical utilities of the patient-specific features learned from the embedding algorithms. The latent patient features obtained via the embedding process enabled direct applications of other machine learning algorithms. Future work will focus on utilising the temporal information within EHR and extending EHR embedding algorithms to develop personalised patient journey predictions.

Distinct Ocular Surface Microbiome in Keratoconus Patients Correlate With Local Immune Dysregulation.

Keratoconus (KC) is characterized by irregular astigmatism along with corneal stromal weakness and is associated with altered immune status. Tissue resident microbiomes are known to influence the immune status in other organs, but such a nexus has not been described in ocular conditions. Therefore, we examined the ocular surface microbiome of patients with KC and correlated it to the immune cell and tear molecular factor profiles. Sixty-two patients with KC and 21 healthy controls underwent corneal topography analysis and eye examination followed by a collection of Schirmer's strip, ocular surface wash, and ocular surface swabs. Microbiomes were analyzed by extracting DNA from the swabs followed by 16S rRNA gene V3-V4 amplicon sequencing and analyzed using QIIME. Fifty-two molecular factors from Schirmer's strip tear extracts and 11 immune cells from ocular wash were measured using multiplex ELISA and flow cytometry. Alpha diversity, linear discriminant analysis effect size (LEfSe), relative abundance and receiver operating characteristic - area under the curve (ROC-AUC) analysis were performed. Unsupervised clustering at the genus level with clinical parameters, soluble factors, and immune cells was performed. Fifty-two phyla/class, 132 order, 283 family, and 718 genera were identified in our cohort. Alpha diversity indices were comparable between patients with KC and the healthy controls. Dominant phyla across groups were Actinobacteria, Proteobacteria, Firmicutes, and Bacteroidetes. Alphaproteobacteria increased in KC eyes whereas Actinobacteria, Firmicutes_Bacilli reduced compared with the healthy controls. We found a significant positive correlation of Microbacterium, Cutibacterium, and Brevundimonas genera abundance with keratometry and corneal thickness. Levels of IL-21, IL-9, Fractalkine, and VEGF positively correlated with Tetrasphaera (P < 0.05). β2-microglobulin and CD66bhigh cells correlated with Bacteroides (P < 0.05). CD45+ cells correlated with Escherichia_Shigella (P < 0.02). We discovered a unique microbiome signature of KC which correlated to disease grades and secreted molecular factors and immune cells. Therefore, the altered microbiome on the ocular surface may drive immune dysregulation in KC and provide scope for potential interventions in the future.

An Integrated Approach Utilizing Single-Cell and Bulk RNA-Sequencing for the Identification of a Mitophagy-Associated Genes Signature: Implications for Prognostication and Therapeutic Stratification in Prostate Cancer

Introduction: Prostate cancer, notably prostate adenocarcinoma (PARD), has high incidence and mortality rates. Although typically resistant to immunotherapy, recent studies have found immune targets for prostate cancer. Stratifying patients by molecular subtypes may identify those who could benefit from immunotherapy. Methods: We used single-cell and bulk RNA sequencing data from GEO and TCGA databases. We characterized the tumor microenvironment at the single-cell level, analyzing cell interactions and identifying fibroblasts linked to mitophagy. Target genes were narrowed down at the bulk transcriptome level to construct a PARD prognosis prediction nomogram. Unsupervised consensus clustering classified PARD into subtypes, analyzing differences in clinical features, immune infiltration, and immunotherapy. Furthermore, the cellular functions of the genes of interest were verified in vitro. Results: We identified ten cell types and 160 mitophagy-related single-cell differentially expressed genes (MR-scDEGs). Strong interactions were observed between fibroblasts, endothelial cells, CD8+ T cells, and NK cells. Fibroblasts linked to mitophagy were divided into six subtypes. Intersection of DEGs from three bulk datasets with MR-scDEGs identified 26 key genes clustered into two subgroups. COX regression analysis identified seven prognostic key genes, enabling a prognostic nomogram model. High and low-risk groups showed significant differences in clinical features, immune infiltration, immunotherapy, and drug sensitivity. In prostate cancer cell lines, CAV1, PALLD, and ITGB8 are upregulated, while CLDN7 is downregulated. Knockdown of PALLD significantly inhibits the proliferation and colony-forming ability of PC3 and DU145 cells, suggesting the important roles of this gene in prostate cancer progression. Conclusions: This study analyzed mitophagy-related genes in PARD, predicting prognosis and aiding in subtype identification and immunotherapy response analysis. This approach offers new strategies for treating prostate cancer with specific molecular subtypes and helps develop potential biomarkers for personalized medicine strategies.

Single-cell analyses of intestinal epithelium reveal the dysregulation of gut immune microenvironment in systemic lupus erythematosus

BackgroundThe small intestine harbors a rich array of intestinal intraepithelial lymphocytes (IELs) that interact with structural cells to collectively sustain gut immune homeostasis. Dysregulation of gut immune homeostasis was implicated in the pathogenesis of multiple autoimmune diseases, however, whether this homeostasis is disrupted in a lupus autoimmune background remains unclear.MethodsWe performed single-cell RNA sequencing (scRNA-seq) analyses to elucidate immune and structural milieu in the intestinal epithelium of MRL/Lpr lupus mice (Lpr mice) and MRL/Mpj control mice (Mpj mice). Comprehensive analyses including unsupervised clustering, trajectories, and cellular communication were performed. The primary findings from scRNA-seq were further validated by quantitative polymerase chain reaction (qPCR), flow cytometry, and in vivo experiments including selenium supplementation.ResultsWe observed a significant reduction in CD8αα + IELs, accompanied by a marked increase in CD8αβ + IELs in Lpr mice. Additionally, subsets of CD8 + IELs exhibiting significantly enhanced effector functions were found to be markedly enriched in Lpr mice. Intercellular communication patterns within intestinal epithelial immune and structural cells were found to be specifically altered in Lpr mice. Moreover, scRNA-seq revealed significantly decreased intestinal TCRγδ T cells (γδT) associated with reduced aryl-hydrocarbon receptor repressor (AHRR) expression and subsequent oxidative stress and ferroptosis in Lpr mice. Antioxidant selenium effectively reversed the loss of γδT in Lpr mice, improved the gut barrier, and alleviated lupus symptoms.ConclusionsOur high-resolution single-cell atlas enhances the understanding of the immune and structural milieu of intestinal epithelium in lupus and provides new insights into lupus pathogenesis mediated by intestinal immune dysregulation.

An efficient heuristic for geometric analysis of cell deformations.

Sickle cell disease causes erythrocytes to become sickle-shaped, affecting their movement in the bloodstream and reducing oxygen delivery. It has a high global prevalence and places a significant burden on healthcare systems, especially in resource-limited regions. Automated classification of sickle cells in blood images is crucial, allowing the specialist to reduce the effort required and avoid errors when quantifying the deformed cells and assessing the severity of a crisis. Recent studies have proposed various erythrocyte representation and classification methods (Jennifer et al., 2023 [1]). Since classification depends solely on cell shape, a suitable approach models erythrocytes as closed planar curves in shape space (Epifanio et al., 2020). This approach employs elastic distances between shapes, which are invariant under rotations, translations, scaling, and reparameterizations, ensuring consistent distance measurements regardless of the curves' position, starting point, or traversal speed. While previous methods exploiting shape space distances had achieved high accuracy, we refined the model by considering the geometric characteristics of healthy and sickled erythrocytes. Our method proposes (1) to employ a fixed parameterization based on the major axis of each cell to compute distances and (2) to align each cell with two templates using this parameterization before computing distances. Aligning shapes to templates before distance computation, a concept successfully applied in areas such as molecular dynamics (Richmond et al., 2004 [2]), and using a fixed parameterization, instead of minimizing distances across all possible parameterizations, simplifies calculations. This strategy achieves 96.03% accuracy rate in both supervised classification and unsupervised clustering. Our method ensures efficient erythrocyte classification, maintaining or improving accuracy over shape space models while significantly reducing computational costs.

Colorectal cancer detection with enhanced precision using a hybrid supervised and unsupervised learning approach

The current work introduces the hybrid ensemble framework for the detection and segmentation of colorectal cancer. This framework will incorporate both supervised classification and unsupervised clustering methods to present more understandable and accurate diagnostic results. The method entails several steps with CNN models: ADa-22 and AD-22, transformer networks, and an SVM classifier, all inbuilt. The CVC ClinicDB dataset supports this process, containing 1650 colonoscopy images classified as polyps or non-polyps. The best performance in the ensembles was done by the AD-22 + Transformer + SVM model, with an AUC of 0.99, a training accuracy of 99.50%, and a testing accuracy of 99.00%. This group also saw a high accuracy of 97.50% for Polyps and 99.30% for Non-Polyps, together with a recall of 97.80% for Polyps and 98.90% for Non-Polyps, hence performing very well in identifying both cancerous and healthy regions. The framework proposed here uses K-means clustering in combination with the visualisation of bounding boxes, thereby improving segmentation and yielding a silhouette score of 0.73 with the best cluster configuration. It discusses how to combine feature interpretation challenges into medical imaging for accurate localization and precise segmentation of malignant regions. A good balance between performance and generalization shall be done by hyperparameter optimization-heavy learning rates; dropout rates and overfitting shall be suppressed effectively. The hybrid schema of this work treats the deficiencies of the previous approaches, such as incorporating CNN-based effective feature extraction, Transformer networks for developing attention mechanisms, and finally the fine decision boundary of the support vector machine. Further, we refine this process via unsupervised clustering for the purpose of enhancing the visualisation of such a procedure. Such a holistic framework, hence, further boosts classification and segmentation results by generating understandable outcomes for more rigorous benchmarking of detecting colorectal cancer and with higher reality towards clinical application feasibility.

Integrative deep immune profiling of the elderly reveals systems-level signatures of aging, sex, smoking, and clinical traits.

Aging increases disease susceptibility and reduces vaccine responsiveness, highlighting the need to better understand the aging immune system and its clinical associations. Studying the human immune system, however, remains challenging due to its complexity and significant inter-individual variability. We conducted an immune profiling study of 550 elderly participants (≥60 years) and 100 young controls (20-40 years) from the RESIST Senior Individuals (SI) cohort. Extensive demographic, clinical, and laboratory data were collected. Multi-color spectral flow cytometry and 48-plex plasma cytokine assays were used for deep immune phenotyping. Data were analyzed using unsupervised clustering and multi-dataset integration approaches. We studied 97 innate and adaptive immune cell populations, revealing intricate age- and sex-related changes in the elderly immune system. Our large sample size allowed detection of even subtle changes in cytokines and immune cell clusters. Integrative analysis combining clinical, laboratory, and immunological data revealed systems-level aging signatures, including shifts in specific immune cell subpopulations and cytokine concentrations (e.g., HGF and CCL27). Additionally, we identified unique immune signatures associated with smoking, obesity, and diseases such as osteoporosis, heart failure, and gout. This study provides one of the most comprehensive immune profiles of elderly individuals, uncovering high-resolution immune changes associated with aging. Our findings highlight clinically relevant immune signatures that enhance our understanding of aging-related diseases and could guide future research into new treatments, offering translational insights into human health and aging. Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy-EXC 2155-project number 390874280.

Assessment of Pine Tree Crown Delineation Algorithms on UAV Data: From K-Means Clustering to CNN Segmentation

The crown area is a critical metric for evaluating tree growth and supporting various ecological and forestry analyses. This study compares three approaches, i.e., unsupervised clustering, region-based, and deep learning, to estimate the crown area of Pinus eldarica Medw. using UAV-acquired RGB imagery (2 cm ground sampling distance) and high-density point clouds (1.27 points/cm3). The first approach applied unsupervised clustering techniques, such as Mean-shift and K-means, to directly estimate crown areas, bypassing tree top detection. The second employed a region-based approach, using Template Matching and Local Maxima (LM) for tree top identification, followed by Marker-Controlled Watershed (MCW) and Seeded Region Growing for crown delineation. The third approach utilized a Convolutional Neural Network (CNN) that integrated Digital Surface Model layers with the Visible Atmospheric Resistance Index for enhanced segmentation. The results were compared against field measurements and manual digitization. The findings reveal that CNN and MCW with LM were the most effective, particularly for small and large trees, though performance decreased for medium-sized crowns. CNN provided the most accurate results overall, with a relative root mean square error (RRMSE) of 8.85%, a Nash–Sutcliffe Efficiency (NSE) of 0.97, and a bias score (BS) of 1.00. The CNN crown area estimates showed strong correlations (R2 = 0.83, 0.62, and 0.94 for small, medium, and large trees, respectively) with manually digitized references. This study underscores the value of advanced CNN techniques for precise crown area and shape estimation, highlighting the need for future research to refine algorithms for improved handling of crown size variability.

Identification of PANoptosis associated lncRNAs associated with clinical prognosis and immune infiltration microenvironment in colon adenocarcinoma

Early diagnosis and disease management based on risk stratification have a very positive impact on colon adenocarcinoma (COAD) prognosis. It is of positive significance to further explore risk stratification of COAD patients and identify predictive molecular biomarkers. PANoptosis is defined as a form of inflammatory cell death regulated by PANoptosome, with common features of pyroptosis, apoptosis and necroptosis. The role of PANoptosis in COAD has not been fully studied. In this study, we analyzed significant differences in the expression of PANoptosis-related gene (PRG) features in COAD. Subsequently, the PANoptosis associated lncRNAs (PALs) associated with PRGs were analyzed by LASSO algorithm and multivariate Cox analysis, and PALs related to the prognosis of COAD were selected. Based on the expression patterns of prognostic PAL features, we performed unsupervised consensus cluster analysis to categorize COAD samples into distinct PAL molecular subtypes and investigate their associated immune infiltration characteristics. We subsequently constructed PAL score model based on prognostic characteristics and verified its independent prognostic value for COAD. The nomogram diagnostic model was established to confirm the prognostic value of PAL scoring system again. Pathway enrichment analysis, somatic mutation profiling, and drug sensitivity analysis were employed to comprehensively assess the clinical value of the PAL score. Additionally, qRT-PCR was used to further validate the abnormal expression of the selected targets in COAD. Our results provide a new idea for clinical risk stratification and new evidence for the role of PANoptosis in COAD.

Classification of Fibro-osseous Tumors in the Craniofacial Bones using DNA Methylation and Copy Number Alterations.

Fibro-osseous tumors of the craniofacial bones are a heterogeneous group of lesions comprising cemento-osseous dysplasia (COD), cemento-ossifying fibroma (COF), juvenile trabecular ossifying fibroma (JTOF), psammomatoid ossifying fibroma (PsOF), fibrous dysplasia (FD), and low-grade osteosarcoma (LGOS) with overlapping clinicopathological features. However, their clinical behavior and treatment differ significantly, underlining the need for accurate diagnosis. Molecular diagnostic markers exist for subsets of these tumors, including GNAS mutations in FD, SATB2 fusions in PsOF, mutations involving the RAS-MAPK signaling pathway in COD, and MDM2 amplification in LGOS. Since DNA methylation and copy number profiling are well established for the classification of central nervous system tumors, our aim was to investigate whether this tool might be used as well for classifying fibro-osseous tumors in the craniofacial bones. We collected a well-characterized, multicenter cohort with available molecular data, including COD (n = 20), COF (n = 13), JTOF (n = 10), PsOF (n = 25), FD (n = 23), LGOS (n = 4), and high-grade osteosarcoma (HGOS; n = 11). Genome-wide DNA methylation and copy number variation data were generated using the Illumina Infinium Methylation EPIC array interrogating >850 000 CpG sites. DNA methylation profiling yielded evaluable results in 73/106 tumors, including 6 CODs, 12 COFs, 6 JTOFs, 19 PsOFs, 18 FDs, 2 LGOSs, and 10 HGOSs. Unsupervised clustering and dimensionality reduction (Uniform Manifold Approximation and Projection) revealed that FD, extragnatic PsOF, and HGOS formed distinct clusters. Surprisingly, COD, COF, JTOF, and mandibular PsOF clustered together, apart from other craniofacial bone tumors. LGOS did not form a distinct cluster, likely due to the low number of cases. Copy number analysis revealed that FD, COD, COF, JTOF, and PsOF were typically characterized by flat copy number profiles compared to LGOS with gains of chromosome 12 and HGOS with multiple heterogeneous copy number alterations. In conclusion, using DNA methylation and copy number profiles, benign fibro-osseous tumors can be separated from low-grade and high-grade osteosarcomas in the craniofacial bones, which is of diagnostic value in challenging cases with overlapping clinicopathological features.

Informing ICU Digital Twins: Dynamic Assessment of Cardiorespiratory Failure Trajectories in Patients with Sepsis.

Understanding clinical trajectories of sepsis patients is crucial for prognostication, resource planning, and to inform digital twin models of critical illness. This study aims to identify common clinical trajectories based on dynamic assessment of cardiorespiratory support using a validated electronic health record data that covers retrospective cohort of 19,177 patients with sepsis admitted to ICUs of Mayo Clinic Hospitals over eight-year period. Patient trajectories were modeled from ICU admission up to 14 days using an unsupervised machine learning two-stage clustering method based on cardiorespiratory support in ICU and hospital discharge status. Of 19,177 patients, 42% were female with a median age of 65 (IQR, 55-76) years, APACHE III score of 70 (IQR, 56-87), hospital length of stay (LOS) of 7 (IQR, 4-12) days, and ICU LOS of 2 (IQR, 1-4) days. Four distinct trajectories were identified: fast recovery (27% with a mortality rate of 3.5% and median hospital LOS of 3 (IQR, 2-15) days), slow recovery (62% with a mortality rate of 3.6% and hospital LOS of 8 (IQR, 6-13) days), fast decline (4% with a mortality rate of 99.7% and hospital LOS of 1 (IQR, 0-1) day), and delayed decline (7% with a mortality rate of 97.9% and hospital LOS of 5 (IQR, 3-8) days). Distinct trajectories remained robust and were distinguished by Charlston comorbidity index, Apache III scores, day 1 and day 3 SOFA (p < 0.001 ANOVA). These findings provide a foundation for developing prediction models and digital twin decision support tools, improving both shared decision-making and resource planning.

Deep unsupervised clustering for prostate auto-segmentation with and without hydrogel spacer

Abstract Introduction. Clinical datasets for training deep learning (DL) models often exhibit high levels of heterogeneity due to differences such as patient characteristics, new medical techniques, and physician preferences. In recent years, hydrogel spacers have been used in some prostate cancer patients receiving radiotherapy to separate the prostate and the rectum to better spare the rectum while achieving adequate dose coverage on the prostate. However, this substantially affects the computed tomography image appearance, which downstream reduced the contouring accuracy of auto-segmentation algorithms. This leads to highly heterogeneous dataset. Methods. To address this issue, we propose to identify underlying clusters within the dataset and use the cluster labels for segmentation. We collected a clinical dataset of 909 patients, including those with two types of hydrogel spacers and those without. First, we trained a DL model to locate the prostate and limit our field of view to the local area surrounding the prostate and rectum. We then used Uniform Manifold Approximation and Projection (UMAP) for dimensionality reduction and employed k-means clustering to assign each patient to a cluster. To leverage this clustered data, we propose a text-guided segmentation model, contrastive language and image pre-training (CLIP)-UNet, which encodes the cluster information using a text encoder and combines the encoded text information with image features for segmentation. Results. The UMAP results indicated up to three clusters within the dataset. CLIP-UNet with cluster information achieved a Dice score of 86.2% compared to 84.4% from the baseline UNet. Additionally, CLIP-UNet outperforms other state-of-the-art models with or without cluster information. Conclusion. Automatic clustering assisted by DL can reveal hidden data clusters in clinical datasets, and CLIP-UNet effectively utilizes clustered labels and achieves higher performance.

Automated Microfluidic Platform for High‐Throughput Biosensor Development

AbstractBiorecognition elements immobilized into nanopores have transformed point‐of‐care (POC) diagnostics by converting molecular interactions into electrical and fluorescent signals.This study introduces Bio‐Sensei, a high‐throughput screening (HTS) microfluidic platform based on nanopore biosensing. Integrating a robotic sampler, electrochemical, and fluorescence setup, Bio‐Sensei operates as an Internet of Things (IoT) platform with integrated data analysis. The platform's utility is demonstrated on functionalized with an amino terminal Cu(II)‐ and Ni(II)‐binding (ATCUN) peptide ion track‐etched membrane. Automated testing atchieves a significantly higher F‐stat value than the critical treshold, while unsupervised clustering reveals optimalnanopores pore size. The biosensordemonstrates remarkable stability, selectivity, and sensitivity with detection limits of 10−6 using fluorescence and 10−15 Musing cyclic voltammetry measurements. Combining these methods enhances machine learning models for Cu2+ concentration prediction, achieving receiver operating characteristic area under the curve values exceeding 95%.

High-parameter immunophenotyping reveals distinct immune cell profiles in pruritic dogs and cats

IntroductionImmunophenotyping is a powerful tool for grading disease severity, aiding in diagnosis, predicting clinical response, and guiding the development of novel therapeutics.MethodsThis pilot study employs high parameter immunophenotyping panels (15 markers for dog, 12 for cat) and leverages unsupervised clustering to identify immune cell populations. Our analysis uses machine learning and statistical algorithms to perform unsupervised clustering, multiple visualizations, and statistical analysis of high parameter flow cytometry data. This method reduces user bias and precisely identifies cell populations, demonstrating its potential to detect variations and differentiate populations effectively. To enhance our understanding of cat and dog biology and test the unsupervised clustering approach on real-world samples, we performed in-depth profiling of immune cell populations in blood collected from client-owned and laboratory animals [dogs (n = 55) and cats (n = 68)]. These animals were categorized based on pruritic behavior or routine check-ups (non-pruritic controls).ResultsUnsupervised clustering revealed various immune cell populations, including T-cell subsets distinguished by CD62L expression and distinct monocyte subsets. Notably, there were significant differences in monocyte subsets between pruritic and non-pruritic animals. Pruritic dogs and cats showed significant shifts in CD62LHi T-cell subsets compared to non-pruritic controls, with opposite trends observed between pruritic cats and dogs.DiscussionThese findings underscore the importance of advancing veterinary immunophenotyping, expanding our knowledge about marker expression on circulating immune cells and driving progress in understanding veterinary-specific biology and uncovering new insights into various conditions and diseases.

Identifying Patient Subgroups in the Heterogeneous Chronic Pain Population Using Cluster Analysis.

Chronic pain is an ill-defined disease with complex biopsychosocial aspects, posing treatment challenges. We hypothesized that treatment failure results, at least partly, from limited understanding of diverse patient subgroups. We aimed to identify subgroups using psychological variables, allowing for more tailored interventions. In a retrospective cohort study, we extracted patient-reported data from two Dutch tertiary multidisciplinary outpatient pain clinics (2018-2023) for unsupervised hierarchical clustering. Clusters were defined by anxiety, depression, pain catastrophizing, and kinesiophobia. Sociodemographics, pain characteristics, diagnosis, lifestyle, health-related quality of life and treatment efficacy were compared among clusters. A prediction model was built utilizing a minimum set of questions to reliably assess cluster allocation. Among 5,466 patients with chronic pain, three clusters emerged. Cluster 1 (n=750) was characterized by high psychological burden, low health-related quality of life, lower educational levels and employment rates, and more smoking. Cluster 2 (n=1,795) showed low psychological burden, intermediate health-related quality of life, higher educational levels and employment rates, and more alcohol consumption. Cluster 3 (n=2,909) showed intermediate features. Pain reduction following treatment was least in cluster 1 (28.6% after capsaicin patch, 18.2% after multidisciplinary treatment), compared to >50% for both treatments in clusters 2 and 3. A model incorporating 15 psychometric questions reliably predicted cluster allocation. In conclusion, our study identified distinct chronic pain patient clusters through 15 psychological questions, revealing one cluster with notably poorer response to conventional treatment. Our prediction model, integrated in a web-based tool, may help clinicians improve treatment by allowing patient-subgroup targeted therapy according to cluster allocation. PERSPECTIVE: Hierarchical clustering of chronic pain patients identified three subgroups with similar pain intensity and diagnoses but distinct psychosocial traits. One group with higher psychological burden showed poorer treatment outcomes. A web-based tool using this model could help clinicians tailor therapies by matching interventions to specific patient subgroups for improved outcomes.

P0616 ViT-based deep learning and unsupervised clustering analysis in Crohn’s Disease based on body composition to identify distinct phenogroups and predict the effectiveness of anti-TNF therapy

Abstract Background Anti-TNF therapy is considered as a crucial treatment in Crohn’s disease (CD). However, it has a primary nonresponse rate in up to 30% of patients and secondary loss of response rate of about 5% per patient-year with unknown reason. Herein, we aimed to establish an artificial intelligence algorithm to identify patients with beneficial response and predict the effectiveness of anti-TNF theray based on body composition. Methods Between July 2020 and August 2023, 105 CD patients who underwent anti-TNF therapy from Shanghai Ninth people’s Hospital were enrolled in this study. An unsupervised clustering analysis (K-means clustering) was utilized to classify subjects into distinct groups based on body composition parameters before anti-TNF therapy. We analyzed the relationship between different phenogroups and the response to anti-TNF therapy. Besides, radiomics features were extracted from subcutaneous fat (SF), visceral fat (VF), and skeletal muscle (SM) at the level of L3/L4 based on CTE. A deep learning model based on Vision Transformer (ViT) was established to predict the response of anti-TNF therapy. Results Our analysis identified three phenogroups: Cluster 1 (PhAhigh/BCMhigh/BMChigh/SMIhigh/BMIhigh/proteinhigh/IWhigh/EWhigh), Cluster 2 (PhAlow/BCMlow/BMClow/SMIlow/VFAhigh/BFPhigh/proteinlow/IWlow/EWlow) and Cluster 3 (VFAlow/BMIlow/BFPlow). Patients in Cluster 1 showed best response of anti-TNF thrapy (p = 0.023). Among patients in Cluster 3, those with high nutritional risks showed significantly earlier loss of response than those with low nutritional risks (p = 0.041). The prediction model achieved satisfactory accuracy in predicting the effectiveness of anti-TNF therapy (SF: AUC = 0.700; VF: AUC = 0.745; SM: AUC = 0.785) in the validation cohort. Conclusion Patients with good nutritional status have a better response to anti-TNF therapy in CD. Earlier loss of response occurred in patients with high nutritional risks, which may be improved by nutritional support. Subcutaneous fat, visceral fat, and skeletal muscle at the level of L3/L4 showed a satisfactory performance in predicting the response to anti-TNF therapy.

P0030 Multiomic Analysis Reveals Three Distinct subtypes within Perianal Crohn’s Disease, Independent of Concomitant Proctitis

Abstract Background Perianal fistulisation (pCD) affects ~20% of Crohn’s disease (CD) patients and often requires combined medical and surgical management. Despite significant progress in management of luminal CD, pCD continues to pose a major unmet need, emphasizing the importance of understanding the molecular and environmental drivers and to discover new therapeutic targets. Methods Paired biopsies from the fistula tract and rectal orifice of 81 pCD patients underwent 16S rRNA and RNA sequencing, generating multiomic profiles for both tissues. We applied Multiomics Factor Analysis (MOFA), to identify key molecular signatures, followed by hierarchical clustering1. Gene module scores were derived using Seurat to visualize expression patterns on a public single-cell RNA seq (scRNA-seq) dataset. Results The MOFA model showed minimal contribution from the microbiota but accounted for over 80% of transcriptomic variance. Unsupervised clustering based on MOFA identified three distinct patient clusters (C1, C2, C3) with unique molecular phenotypes, independent of concomitant or prior proctitis (p=1.0) (Fig 1A). C3 was associated with positive end of the primary axis of variation, showing a keratinization signature in the fistula, without signs of epithelial-mesenchymal transition (EMT) or rectal keratinization (Fig 1B). In contrast, C1 and C2 were linked to the opposite extreme of this axis, characterized by EMT signatures in the fistula and an absence of keratinization. The rectal profiles differed between these two clusters: C1 lacked keratinization in the rectum, while C2 exhibited concurrent keratinization (Fig 1B). The opposing pattern along the primary axis suggests that keratinization and EMT represent distinct, mutually exclusive processes in fistula pathology. Analysis of publicly available scRNA-seq data from psoriasis patients (a keratinization-hallmark disease) revealed significant upregulation of the pCD-specific keratinization gene signature in lesional keratinocytes compared to non-lesional samples, indicating a psoriasis-like keratinization process in pCD2 (Fig 2)2. Finally, microbial profiling showed no significant associations between bacterial genera and any of the identified clusters in either rectum or fistula (adj. p &amp;gt; 0.05) Conclusion Our analysis identified three distinct patient clusters in pCD, revealing diverging keratinization and EMT phenotypes in fistula. This stratification was independent of proctitis and there was little to no contribution of the microbiota. Comparing the expression profiles with the ones observed in patients with psoriasis suggests a psoriasis-like keratinization process in a subset of patients. These findings uncover potential novel mechanisms in pCD pathogenesis.