Abstract Single-cell RNA sequencing allows defining cellular identities based on transcriptional similarity using unsupervised clustering. However, a single clustering resolution may not yield groups of cells that represent both broad, well-defined populations and smaller subpopulations simultaneously. Therefore, when cell identities are not known prior to sequencing, robust comparison and annotation of inferred de novo clusters remains a challenge. Here, we introduce CANTAO, in which we propose the average overlap metric to define the distance between single-cell clusters by comparing ranked lists of differentially expressed genes in a top-weighted manner. We benchmark CANTAO in truth-known datasets comprised of similar yet distinct cell populations and show that evaluating clusters with average overlap results in a consistent, precise, and biologically meaningful recapitulation of true cell identities. We then analyze unsorted mouse thymocytes and characterize stages of T-cell development in the thymus, including minor populations of double-negative (CD4-CD8-) T cells that are difficult to confidently detect among unsorted single cells. We demonstrate that CANTAO enables robust, reproducible characterization of single-cell data and clarifies biological interpretation of underlying identities in homogeneous populations.

Abstract Infections are recognised triggers for several neuroinflammatory disorders. The COVID-19 pandemic’s nonpharmaceutical interventions sharply curtailed pathogen exposure, creating a natural experiment to test infection-disease links. Using Japan’s National Claims Database, we first validated the nationwide decline with two strictly infection-dependent conditions—epidemic keratoconjunctivitis and influenza-associated encephalopathy—whose monthly incidences fell by >70% after April 2020. Next, we applied an interrupted time-series design, a causal-inference method for longitudinal data, to nine immune-mediated inflammatory diseases. Unsupervised clustering of model-derived level and slope changes identified three data-driven clusters. The first cluster, comprising Guillain–Barré syndrome and acute disseminated encephalomyelitis, showed large, statistically significant level drops (p < 0.001), particularly in women, consistent with infection-susceptible pathophysiology. The second cluster, including myasthenia gravis and optic neuritis, exhibited transient declines followed by significant positive post-intervention slopes (p < 0.001), suggesting deferred diagnosis, treatment interruption, or immune rebound. The third cluster, consisting of sarcoidosis, neuromyelitis optica, multiple sclerosis, Vogt–Koyanagi–Harada disease, and Behçet’s disease, remained stable, suggesting limited or complex infectious links. These data-driven trajectories mirror clinical pathophysiology and demonstrate that reduced pathogen exposure affects neuroinflammatory disease onset to varying degrees. This framework supports infection-related risk stratification, preventive strategies, and continuity planning in neuroimmunology practice.

This study aimed to investigate the role of stemness in acute myeloid leukemia (AML), stratify patients into subtypes based on stemness-associated signatures, and explore their prognostic implications as well as potential therapeutic vulnerabilities. To investigate the diagnostic and prognostic implications of stemness in AML, we integrated and analyzed comprehensive datasets from the TCGA, GEO, and cBioPorta databases. Initially, stemness and immune scores were calculated using transcriptomic data from patients in the TCGA-LAML training cohort. Unsupervised clustering was then employed to identify two distinct stemness subgroups. Survival analyses were performed for patients in these subgroups using two independent validation cohorts, GSE106291 and OSHU-AML. Furthermore, four complementary machine learning algorithms were employed to evaluate feature importance and identify key stemness-associated genes. Finally, comparative analyses were conducted between the two stemness subgroups to evaluate differences in clinical characteristics, immune cell infiltration patterns, immune scores, expression of immune checkpoint molecules, and predicted responses to therapeutic agents. Our analysis revealed that patients in stemness subgroup II exhibited poorer prognoses, however, treatment with PD-1 inhibitors demonstrated significant efficacy in this subgroup. Conversely, patients in stemness subgroup I displayed enhanced sensitivity to conventional chemotherapies, including Cytarabine, Methotrexate, and Etoposide, compared to those subgroup II. A striking divergence in mutation profiles was observed between the two subgroups, suggesting the engagement of distinct biological processes. Additionally, we identified eight stemness-related genes as potential biomarkers for therapeutic stratification. In conclusion, we have established two distinct stemness subgroups within AML based on stemness scores, and highlighted their differential responses to immunotherapy and conventional treatments. These findings offer novel insights into clinical stratification and therapeutic targeting in AML, paving the way for more personalized treatment approaches.

Background The neonatal gut microbiome plays a critical role in infant health through the production of short-chain fatty acids (SCFAs). However, the organization of SCFAs-producing microbial communities in neonates remains poorly characterized. This study applied unsupervised clustering and machine learning to classify microbial subgroups associated with SCFAs production, providing insight into their composition and metabolic potential. Methods This study recruited 71 mother-infant pairs from Kangwon National University Hospital and Bundang CHA Hospital, collecting meconium samples within five days postpartum. Microbial diversity was analyzed by 16S rRNA gene sequencing (V3–V4 region) at the genus level, and SCFAs were quantified from the same samples. To identify functionally distinct microbial subgroups, K-Means, Agglomerative, Spectral, and Gaussian Mixture Model clustering were applied. Clustering validity was assessed using Silhouette Score, Calinski-Harabasz Index, Davies-Bouldin Index, and Prediction Strength Validation, with t-distributed Stochastic Neighbor Embedding (t-SNE) visualization to evaluate cluster separation. SCFAs distributions across clusters were compared, while random forest and logistic regression models were used to classify SCFAs-associated microbial clusters through Receiver Operating Characteristic curves (ROC). Results The clustering analysis identified distinct microbial subgroups linked to SCFAs production, with Agglomerative clustering outperforming K-Means in capturing functionally relevant structures. Cluster 1 had higher SCFAs levels, enriched in Bacteroides , Prevotella , and Enterococcus , while Cluster 2 exhibited lower SCFAs concentrations with a more heterogeneous composition. The introduction of a third cluster in multi-class analysis revealed an intermediate metabolic profile, suggesting a continuum in microbial metabolic function. Classification analysis confirmed random forest model superiority, achieving ROC score of 91.05% (Agglomerative) and 87.74% (K-Means) in binary classification, and 92.98% (Agglomerative) and 89.84% (K-Means) in multi-class classification, demonstrating RF’s strong predictive ability for SCFAs-based clusters. Conclusion Unsupervised clustering combined with classification analysis effectively predict SCFAs-associated subgroups and paving the way for future research on longitudinal tracking and functional genomic integration in early-life metabolic health.

Abstract Comparative genomics enables scientific exploration of genetic variations across species, enabling insights into human diseases that are indiscernible from analysis of human data alone. Dogs, in particular, are an excellent model organism in comparative oncology because they develop spontaneous tumors that closely resemble human cancers. Many recent studies have investigated parallels between dog and human cancers in a single tumor type. However, to date, no study has systematically quantified the effectiveness of the dog model as a pan-cancer representation of adult and pediatric cancers. To address this gap, we performed a pan-cancer RNA-seq analysis of human and dog cancers spanning 6,048 samples and 11 tumor types, including 987 dog tumor samples from Cahill et al., 4,272 adult samples from TCGA, and 789 pediatric samples from the Treehouse compendium. We analyzed 10 adult and 5 pediatric cancers with at least 10 tumor samples for each species. Using a combination of exploratory data analysis, differential expression and enrichment analysis, unsupervised clustering, and supervised modeling, we identified shared expression patterns across species. We developed a weighted formula incorporating results from these analyses to quantitatively assess the translational relevance of the dog model for each cancer and applied it across cancer types. Our formula ranked dog cancers by transcriptional similarity to human cancers and revealed which models best capture tumor heterogeneity. The cancer-type specific results from this study match those from individual cancer studies, suggesting that our formula effectively identifies the tumor types where dog models best reflect human cancers. Our work introduces a unified, quantitative framework for evaluating cross-species cancer models through combining multiple analysis methods to support preclinical studies and accelerate therapy development. This scalable approach enables fast and accurate identification of model systems to study rare human cancers. As genome sequencing increases, this automated approach minimizes the manual effort required to identify suitable model organisms across the tree of life and opens new opportunities for comparative oncology studies. Citation Format: Geesa Daluwatumulle, Leslie Smith, Nathan Glen, James Cahill, Kiley Graim. Quantifying the translational relevance of naturally occurring dog cancers as models of adult and pediatric tumors [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Cancer Evolution: The Dynamics of Progression and Persistence; 2025 Dec 4-6; Albuquerque, NM. Philadelphia (PA): AACR; Cancer Res 2025;85(23_Suppl):Abstract nr B029.

Efferocytosis, the phagocytic clearance of apoptotic cells, exerts dual anti-inflammatory and pro-tumorigenic effects, while existing studies have not well-elucidated the underlying mechanisms in Lung adenocarcinoma (LUAD). This study aims at identifying prognostic genes related to efferocytosis and potential therapeutic targets in LUAD. LUAD-related gene expression data sets were obtained from UCSC Xena and GEO databases, while efferocytosis-related genes (ERGs) were sourced from the GeneCards database. Unsupervised consensus clustering analysis assisted in classifying LUAD patients into two distinct subgroups, after which differential gene expression was analyzed for identifying differentially expressed genes (DEGs). Subsequently, a four-gene prognostic model was built by virtue of LASSO-Cox regression and further validated by Kaplan-Meier (KM) analysis and receiver operating characteristic (ROC) curve assessment. Analysis on single-cell RNA sequencing (scRNA-seq) data engaged in investigating cell subpopulations associated with efferocytosis and to pinpoint key genes involved. In vitro experiments were conducted to ascertain the functional importance of the selected key gene in LUAD progression. A prognostic signature incorporating four genes was successfully constructed for LUAD, and risk scores were calculated for categorizing patients into low- and high-risk groups. Notably, the two risk groups presented different survival outcomes and immune cell infiltration (ICI). Analysis of scRNA-seq data revealed macrophages as the primary efferocytosis-related cell type in LUAD and pinpointed LDHA as a key regulatory gene. Pseudotime trajectory analysis demonstrated a progressive increase in LDHA expression and efferocytosis-related activity as macrophages differentiated into tumor-associated macrophages (TAMs). Moreover, in vitro experimental data showed that LDHA silencing in macrophages facilitated polarization toward an M1 phenotype and simultaneously inhibited M2 polarization. In addition, co-culturing LUAD cells with LDHA-silenced M2 macrophages substantially reduced cancer cell proliferation, migration, and invasion. Cell communication analyses suggested that LDHA promotes M2-type TAMs by modulating the SPP1 pathway between macrophages and cancer-associated fibroblasts (CAFs), activating ECM receptors and downstream effectors. We constructed an efferocytosis-related prognostic gene model for LUAD. Furthermore, LDHA may target M2 macrophages, highlighting its potential as a therapeutic target in LUAD.

Abstract Leveraging unsupervised clustering to capture transcriptomic variation and map temporal dynamics in disease progression provides a framework for linking molecular biology to heterogeneous clinical phenotypes. Radical prostatectomy (RP) or radiation therapy are standard treatments for localized prostate cancer (PCa), yet up to ∼15-30% of men experience a rise in blood prostate-specific antigen (PSA) levels (termed biochemical recurrence; BCR) after definitive local therapies for curative intent, reflecting potential disease recurrence. BCR represents a spectrum of disease states with varying implications for subsequent radiographic and/or clinical progression, reflecting the complex underlying tumor biology of PCa. The time until a patient experiences a post-RP rise in PSA—termed time to biochemical recurrence (tBCR)—is an important marker as it can provide insights into risk for clinical progression. The molecular drivers of the BCR disease state are not clearly defined, which is necessary to better understand drivers of recurrence and optimize management of this population of patients. Our initial study has examined transcriptomic data from 501 patients in the TCGA Prostate Adenocarcinoma dataset who underwent RP, of which 74 experienced BCR. Patients with BCR were stratified into quartiles by “Days to BCR” and their gene expression profiles were analyzed using Gene Set Variation Analysis, K-means Clustering, Principal Component Analysis (PCA), and Over-Representation Analysis (ORA). Further, disease free survival (DFS) and overall survival (OS) were assessed to determine clinical relevance of identified disease signatures. Notably, K-means clustering identified a distinct 95-gene cluster showing a linear increase in gene expression that stratifies early from late and never recurrent groups. PCA was subsequently used to identify the top 10 contributing genes to this linearity. Overexpression of epithelial splicing regulatory protein 1 (ESRP1), a regulator of epithelial-mesenchymal transition, accounted for the most variance in the principal component and was found to be significantly associated with early recurrence, as well as decreased DFS and OS in both the TCGA and in an independent metastatic prostate cancer dataset. Additionally, ORA identified a 26-gene signature related to immune cells, i.e., tumor-associated macrophages (TAMs), dendritic cells (DCs), and natural killer (NK) cells, with a 6-gene subset common to all three immune cell types. This 6-gene signature was also associated with early recurrence and DFS/OS in two independent datasets. Analysis of clinical genomic, transcriptomic, and outcomes data for localized and metastatic disease confirmed that K-means clustering effectively separated patients by tBCR via unique transcriptomic signatures, identifying signatures that predict poor prognoses in independent PCa datasets, and suggest modulation of EMT processes and perturbation of immune cell populations in patients who experience early biochemical recurrence. Citation Format: Zumar Meher, Safiullah Rifai, Azimullah Rifai, Tausif Khan, Wei Guang, Mohamad Khan, Arif Hussain. Molecular drivers of biochemical recurrence post radical prostatectomy among men with prostate cancer [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Cancer Evolution: The Dynamics of Progression and Persistence; 2025 Dec 4-6; Albuquerque, NM. Philadelphia (PA): AACR; Cancer Res 2025;85(23_Suppl):Abstract nr B036.

Postoperative recurrence remains a major clinical challenge in patients with resectable invasive adenocarcinoma of the lung (IAC). Conventional PET-based parameters and whole-tumor radiomics may insufficiently reflect the spatial heterogeneity relevant to recurrence risk. This study aimed to develop and validate an interpretable radiomics model based on [1⁸F]FDG PET-derived habitat imaging for individualized recurrence risk prediction. This retrospective study included 156 patients with pathologically confirmed IAC who underwent preoperative [1⁸F]FDG PET/CT. Tumors were segmented and subdivided into intratumoral habitats using voxel-wise radiomic features and unsupervised clustering. A combined model was constructed by integrating clinical variables and radiomic features from the most predictive habitat. Model performance was assessed using receiver operating characteristic analysis, calibration curves, and decision curve analysis. Model interpretability was evaluated using Shapley Additive Explanations (SHAP). Prognostic value was assessed using Kaplan-Meier analysis based on disease-free survival (DFS). The habitat-based combined model demonstrated the highest predictive performance, achieving an area under the curve of 0.823 in the test cohort, with good calibration and clinical utility. Stratification based on model output showed significant differences in DFS between high- and low-risk groups, with P < 0.0001 in the training cohort and P = 0.018 in the test cohort. This PET-based habitat radiomics model provides a noninvasive and interpretable tool for preoperative prediction of postoperative recurrence in IAC. By accurately identifying patients at high risk of recurrence and reduced DFS, the model may support risk-adapted decision-making for postoperative management.

Single-cell Raman imaging reveals fructose impairs brown adipocyte differentiation.

Enhancing stroke recovery assessment: A machine learning approach to real-world hand function analysis.

Comparison of QuPath and HALO Platforms for Analysis of the Tumor Microenvironment in Prostate Cancer.

Automated note annotation after bioacoustic classification: Unsupervised clustering of extracted acoustic features improves detection of a cryptic owl

Data-driven classification of packaged products commercialized in Uruguay: Insights for the debate on processed food classification systems.

Elderly-onset systemic sclerosis defines a distinct clinical subset: analysis from the SPRING registry of the Italian Society for Rheumatology.

ABSTRACTLung cancer prognosis varies significantly among patients, highlighting the need for accurate prediction tools. Emerging evidence suggests that the immune microenvironment plays a crucial role in lung cancer progression and treatment response. We collected RNA expression profiles and clinical data of lung cancer patients from TCGA and GEO databases. Differential expression analysis identified 276 lung cancer‐associated genes using strict statistical criteria (logFC > 1, FDR < 0.05). Unsupervised consensus clustering divided patients into ‘lung cancer‐related’ and ‘non‐lung cancer‐related’ subgroups. We evaluated 10 machine learning algorithms and 101 algorithmic combinations for prognostic model development. Single‐cell RNA sequencing data were analysed using Seurat and CellChat to investigate immune cell interactions within the lung cancer microenvironment. Our prognostic model demonstrated excellent predictive performance with AUC values of 0.874, 0.891 and 0.925 at 1, 2 and 3 years, respectively (C‐index = 0.874). Six key immune genes (TLR2, TLR4, CCR7, IL18, TIRAP and FOXP3) showed cell‐type specific expression patterns in the lung cancer microenvironment. Intercellular communication analysis revealed complex signalling networks between B cells, T cells, NK cells and dendritic cells. CIBERSORT and ESTIMATE analyses confirmed significant differences in immune infiltration between high‐risk and low‐risk patients, with distinct patterns of T cell subsets, macrophages and dendritic cells. This study provides a reliable prognostic tool for lung cancer and offers insights into the critical role of the immune microenvironment in lung cancer pathogenesis. Our findings may guide the development of personalised immunotherapy strategies for lung cancer patients.

Acute respiratory distress syndrome (ARDS) is a life-threatening condition with significant heterogeneity in pathophysiology. The integration of pulmonary edema indices, respiratory mechanics, and gas exchange parameters to define subphenotypes in mechanically ventilated patients with ARDS has not yet been investigated. We conducted a post hoc analysis of a prospective observational study with a derivation cohort (n = 111). We applied K-means clustering to identify distinct subphenotypes based on key physiological parameters: pulmonary edema indices, respiratory mechanics, and gas exchange variables. The primary outcome was 28-day mortality. Between-group differences in 28-day mortality were analyzed using the chi-square test. Survival analysis was performed with Kaplan-Meier curves (compared by log-rank test) and multivariable Cox regression to adjust for covariates. Furthermore, we compared the differential responses to prone positioning ventilation among the identified subphenotypes to evaluate its potential interaction effect on mortality. An independent validation cohort (n = 55) was used to confirm the subphenotype classifications and their relationships with clinical outcomes. Unsupervised clustering revealed two distinct subphenotypes. Subphenotype 2, characterized by elevated pulmonary vascular permeability index (PVPI) and ventilation ratio (VR), demonstrated significantly higher 28-day mortality compared to Subphenotype 1 (50.0% vs. 28.2%, p = 0.021). This survival disadvantage was confirmed by Kaplan-Meier analysis (log-rank p = 0.016) and a multivariable Cox regression model (adjusted hazard ratio [HR] 2.263, 95% confidence interval [CI] 1.206-4.245; p = 0.011). Furthermore, a statistically significant interaction was observed between subphenotypes and response to prone positioning for 28-day mortality (p-for-interaction = 0.015). Crucially, the prognostic distinction between subphenotypes and their differential treatment response were consistently replicated in an independent validation cohort. Using unsupervised machine learning, this study identified two distinct ARDS subphenotypes characterized by divergent profiles in pulmonary edema, respiratory mechanics, and gas exchange. These subphenotypes were associated with significantly different clinical outcomes and exhibited a differential response to prone positioning therapy. Future research should prioritize the execution of large-scale, multicenter, randomized controlled trials to validate these findings and advance the clinical implementation of precision medicine in the management of ARDS.

ABSTRACTSpinal cord aging is a critical physiological process that compromises central nervous system (CNS) homeostasis and plasticity. Exercise, as a systemic intervention with broad health benefits, has been shown to delay neurodegeneration and preserve tissue function; however, its impact on dynamic cellular lineage evolution and intercellular communication within the aging spinal cord remains poorly characterized. In this study, we employed single‐nucleus RNA sequencing (snRNA‐seq) to construct a high‐resolution cellular atlas of the mouse spinal cord under young, aged, and aerobic exercise‐intervened conditions. By integrating unsupervised clustering, cell proportion analysis, pseudotime trajectory reconstruction, gene regulatory network (GRN) inference, and intercellular communication mapping, we systematically characterized transcriptional and cellular alterations associated with aging and their modulation through exercise. Aging induced pronounced shifts in cell‐type composition and subpopulation structures, which were partially reversed by exercise intervention. Pseudotime analyses of oligodendrocytes, astrocytes, and microglia revealed that exercise remodeled their differentiation trajectories and restored functional states associated with myelin formation, metabolic homeostasis, and inflammation control. GRN analysis identified several key regulators whose centrality and expression were disrupted during aging but reestablished after exercise, suggesting a recovery of transcriptional network organization. Furthermore, intercellular communication analysis revealed reduced signaling strength and connectivity during aging, particularly within gap junction pathways, which were partially restored by exercise, indicating improved cellular coordination. Together, these findings provide a comprehensive single‐cell landscape of the aging spinal cord and demonstrate that exercise reprograms cellular lineages and regulatory networks, offering mechanistic insights into how it mitigates CNS aging and preserves neural function.

This study aimed to develop and validate a radiologic clustering model using CT imaging features to stratify clear cell renal cell carcinoma (ccRCC) patients by prognosis and identify key imaging predictors of 5-year progression free survival (PFS). This retrospective study included 164 ccRCC patients with multiphase kidney CT and next-generation sequencing (NGS) between September 2003 and October 2024. Qualitative imaging features were extracted, and unsupervised consensus clustering was performed to classify tumors based on radiologic characteristics. A nomogram-based C1 score was derived from features predictive of the high-risk cluster. Model performance was evaluated using C-index and 5-year area under the receiver operating curve (AUC). Genetic alterations and copy number variations (CNVs) were also analyzed for associations with imaging features and survival. Clustering revealed two distinct radiologic subtypes. Cluster C1 characterized by aggressive behavior such as tumor heterogeneity (p = 0.011), exophytic growth pattern (p = 0.002), non-smooth margin (p = 0.019), and renal sinus extension (p = 0.016), and was independently associated with poorer 5-year PFS (p = 0.018). The C1 score demonstrated an AUC of 0.992 for predicting cluster C1 in the test-set. Using a cutoff of 0.75, the model achieved 96.3% sensitivity and 96.4% specificity. For predicting 5-year PFS, the C1 score showed moderate performance (AUC 0.65; C-index 0.65), which improved when combined with nodal/distant metastasis and BAP1 mutation status (AUC 0.71; C-index 0.67). Radiologic clustering using CT features enables non-invasive prognostic stratification of ccRCC. The C1 score derived from this approach may serve as a practical tool to guide surveillance and treatment decisions. Retrospectively registered.

This study investigates the dynamics of environmental transformation in the southeastern basins of Madre de Dios, Peru, by integrating multi-spectral remote sensing data with advanced machine learning methodologies. To capture and quantify land surface changes over time, satellite imagery from Landsat and Sentinel missions was utilized to derive key spectral indices—specifically, the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI). These indices provided critical insights into vegetation health and surface water distribution. To manage the high dimensionality of the spectral data, Principal Component Analysis (PCA) was applied, enabling more efficient data interpretation and visualization. Subsequently, unsupervised K-means clustering was employed to classify land cover types and detect spatial patterns of change without prior labeling. The analysis revealed a significant decline in dense vegetative cover, accompanied by a notable expansion of bare soil and surface water areas. These findings point to accelerating environmental degradation in the region, likely driven by both natural and anthropogenic pressures. The methodological framework adopted in this study demonstrates strong potential for scalable, data-driven environmental monitoring and offers a replicable model for assessing land cover dynamics in other ecologically sensitive regions.

BackgroundCholesterol metabolism (CM) plays a critical role in the progression of colorectal cancer (CRC), yet its molecular and immunological implications remain incompletely understood. Therefore, we aimed to identify CRC subtypes according to CM-related genes and reveal their distinct characteristics.MethodsBased on CM-related genes, we applied unsupervised clustering to classify CRC into two subtypes using transcriptomic data from TCGA and comprehensively compared their transcriptomic, genomic and clinical characteristics. We utilized single-cell RNA sequencing data and classified the samples into two subtypes and investigated the distinctions in the tumor microenvironment (TME) between these subtypes.ResultsTwo distinct CM subtypes were identified: Subtype A, characterized by cholesterol esterification and storage, was associated with inflammatory activation and cellular senescence. This subtype exhibited a poor prognosis and reduced predicted response to chemotherapy and immunotherapy. Tumor cells in Subtype A exhibited characteristics of epithelial-mesenchymal transition and angiogenesis. The TME in Subtype A contained higher infiltration of myeloid cells, fibroblasts, and pericytes, with dominant immunosuppressive tumor-associated macrophages (TAMs), especially TAM_SPP1, which interacted closely with Fibro_IL32, promoting immune exclusion. In contrast, Subtype B was marked by enhanced cholesterol catabolism and regulation. Tumor cells in this subtype displayed features of proliferation and stem-like properties. It showed a more active immune microenvironment with increased plasma cell infiltration and fewer immunosuppressive TAMs. Finally, we constructed a prognostic signature and validated its performance across multiple datasets.ConclusionsThese findings provide comprehensive insights into CM subtypes in CRC, highlighting their clinical significance and potential therapeutic implications.Supplementary InformationThe online version contains supplementary material available at 10.1007/s12672-025-03993-z.