---
title: "CCL26 as a Core Comorbid Gene and Biomarker for Atopic Dermatitis and Allergic Asthma — Machine Le"
id: "frontiers-in-immunology-10-identification-and-experimental-validation-of-ccl26-as-a-core-comorbid-gene-and"
canonical_url: "https://medichelpline.com/clinical-feed/frontiers-in-immunology-10-identification-and-experimental-validation-of-ccl26-as-a-core-comorbid-gene-and"
content_type: "clinical_feed_article"
specialty: "Infectious Disease"
source_name: "Frontiers in Immunology"
source_url: "https://www.frontiersin.org/articles/10.3389/fimmu.2026.1936742"
published_at: "2026-09-09T00:00:00.000Z"
evidence_level: "Journal Feed"
license: "CC-BY-NC-4.0 / Informational Use"
---
# CCL26 as a Core Comorbid Gene and Biomarker for Atopic Dermatitis and Allergic Asthma — Machine Le
## Provenance & Clinical Metadata
- **Canonical URL:** https://medichelpline.com/clinical-feed/frontiers-in-immunology-10-identification-and-experimental-validation-of-ccl26-as-a-core-comorbid-gene-and
- **Specialty:** [Infectious Disease](https://medichelpline.com/clinical-feed/infectious-disease.md)
- **Primary Source:** Frontiers in Immunology
- **Source URL:** [Original Journal Publication](https://www.frontiersin.org/articles/10.3389/fimmu.2026.1936742)
- **Published At:** 2026-09-09T00:00:00.000Z
- **Evidence Rating:** Journal Feed
## Executive GIST (TL;DR)
- The source article title reports identification and experimental validation of **CCL26** as a core comorbid gene and potential biomarker for **atopic dermatitis** and **allergic asthma**, using **machine learning** and **multi-omics** approaches. - The publicly available source text supplied here does not include the article abstract, methods, cohort details, data, specific results, or conclusions; those details were not reported in the provided source. - Because the full study content was not included in the supplied material, critical information such as sample sizes, datasets used, machine-learning models, multi-omics platforms, statistical analyses, validation experiments, effect sizes, and p values are not available for summary. - The absence of study methods and results prevents assessment of diagnostic or prognostic performance metrics for **CCL26**, including sensitivity, specificity, or independent validation status. - Important translational implications (for example, whether **CCL26** could guide therapy selection, risk stratification, or serve as a therapeutic target) were not reported in the provided content and therefore cannot be claimed. - Readers should consult the full published article or its supplementary materials for complete experimental details, data availability, and reproducibility statements before applying these findings clinically or in research planning. - The supplied material permits only a high-level statement of the article’s reported focus; no experimental or quantitative claims can be confirmed from the provided text.
## Clinical Analysis & Structured Key Points
Frontiers | Identification and experimental validation of CCL26 as a core comorbid gene and biomarker for atopic dermatitis and allergic asthma via machine learning and multi-omics analysis ORIGINAL RESEARCH article Front. Immunol. , 09 September 2026 Sec. Cytokines and Soluble Mediators in Immunity Volume 17 - 2026 | https://doi.org/10.3389/fimmu.2026.1936742 Published in Frontiers in Immunology Cytokines and Soluble Mediators in Immunity 7 impact factor 11.3 citescore Part of a Research Topic Chemokines guiding cell migration in health and disease Submission open 876 views 1 articles Editor & Reviewers Edited by A D ANNALISA DEL PRETE Reviewed by L Z Lu Zhang S L Shixiu Liang Outline Figures and Tables Figure 1 View in article Figure 2 View in article Figure 3 View in article Figure 4 View in article Figure 5 View in article Figure 6 View in article Figure 7 View in article Figure 8 View in article Figure 9 View in article Figure 10 View in article Figure 11 View in article Figure 12 View in article ORIGINAL RESEARCH article Front. Immunol. , 09 September 2026 Sec. Cytokines and Soluble Mediators in Immunity Volume 17 - 2026 | https://doi.org/10.3389/fimmu.2026.1936742 Identification and experimental validation of CCL26 as a core comorbid gene and biomarker for atopic dermatitis and allergic asthma via machine learning and multi-omics analysis Y G Yunfan Gu 1 S S Shilei Shi 2 Y R Yuqing Rao 1 S Z Shuang Zhang 2 J Z Jiaqi Zhang 2 W Z Weiming Zhang 3 X Z Xianyu Zeng 1,3 * 1. School of Traditional Chinese Medicine, Hubei University of Chinese Medicine, Wuhan, China 2. First Clinical College, Hubei University of Chinese Medicine, Wuhan, China 3. Department of Dermatology, Traditional Chinese and Western Medicine Hospital of Wuhan, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, Hubei, China See more Article metrics View details Abstract Background: Atopic dermatitis (AD) and allergic asthma (AA) frequently co-occur and are a core component of the “atopic march.” However, the shared molecular mechanisms linking these conditions, as well as the core comorbid genes with diagnostic value, remain poorly understood. Methods: This study integrated differential expression analysis and weighted gene co-expression network analysis (WGCNA) of bulk RNA sequencing data from AD and AA to identify comorbidity-associated genes. Diagnostic models were built using 8 feature selection algorithms and 175 machine-learning combinations, and then validated across multiple independent external cohorts. Single-cell transcriptomic data from AD and AA were integrated to identify pathogenic cell subclusters using the Scissor, scAB, and DEGAS algorithms and to examine the distribution of comorbid genes. Their functional mechanisms were clarified through virtual gene knockouts (Geneformer, CellOracle, and scTenifoldKnk) and cell-cell communication analyses (CellChat and MultiNicheNet), and were validated in situ using 10x Visium (AD) and Xenium (AA) spatial transcriptomics data. Finally, Candidate hub genes were validated in vitro using IL-4 and IL-13-stimulated tissue-specific cellular models. Results: Following differential expression analysis, WGCNA, and machine-learning-based feature selection, CCL26 showed consistent and significant upregulation across all cohorts. A diagnostic model based on CCL26 performed well in the external validation cohort. Single-cell analysis revealed that CCL26 is specifically enriched in pathogenic cell subsets, including inflammatory vascular smooth muscle cells and fibroblasts in AD skin, as well as basal- and secretory-lineage epithelial cells in AA airways. Virtual gene knockout revealed that CCL26 is involved in regulating tissue remodeling and inflammatory pathways. Cell-to-cell communication and spatial transcriptomic analyses further confirmed that CCL26 mediates communication between pathogenic cell subsets and Th2 cells via the CCR3 receptor, leading to the formation of self-amplifying inflammatory microenvironments in the skin and airways. In vitro experiments confirmed that CCL26 expression was significantly elevated in IL-4 and IL-13-stimulated HDF and BEAS-2B cells. Conclusion: This study identifies CCL26 as a core comorbid gene and a potential diagnostic biomarker linking AD and AA. Furthermore, it elucidates the molecular mechanism by which CCL26 drives skin-airway inflammation via the CCL26-CCR3 axis, offering a novel perspective on atopic comorbidities. 1 Introduction Atopic dermatitis (AD) and asthma are two common chronic inflammatory diseases that impose a heavy medical and socioeconomic burden globally. According to the 2019 Global Burden of Disease study estimates, asthma affects approximately 262 million people worldwide and leads to about 455,000 deaths annually ( 1 ); the prevalence of AD is as high as one-fifth during childhood, and the adult prevalence also remains at 3%–5% ( 2 ). Although the number of affected individuals for both diseases continues to rise due to population growth, the age-standardized prevalence rates between 1990 and 2019 have decreased, and both show a bimodal pattern, peaking in childhood (ages 5–9 years) and adulthood ( 3 ). More importantly, there is a clear comorbid relationship between the two, termed the “atopic march,” which typically begins with AD in early childhood and subsequently progresses to asthma and/or allergic rhinitis ( 4 ). A meta-analysis incorporating 39 prospective cohort studies with a total of 458810 participants showed that the relative risk of AD patients developing asthma is 2.16, and the risk increases in a dose-dependent manner with the duration and severity of AD ( 5 ); this dose-response relationship was also confirmed in a United Kingdom electronic health record cohort of approximately 5.5 million patients, and the severity of AD is also directly associated with an elevated risk of asthma exacerbations and hospitalizations ( 6 ). Within the AD population, the comorbid prevalence rates of asthma, rhinitis, or their coexistence are 40.5%, 25.7%, and 14.2%, respectively, and the odds of patients developing another atopic disease are 3 to 4 times higher than in the non-AD population ( 2 ). Notably, asthma is a highly heterogeneous clinical syndrome that can be classified into two major endophenotypes, allergic and non-allergic, based on serum-specific IgE levels and skin prick test results. The impact of AD on the two is not uniform; a 40-year prospective cohort study showed that childhood eczema and rhinitis can predict allergic asthma (AA) in adulthood but not non-allergic asthma ( 7 ). This suggests that the causal chain of the “atopic march” is essentially driven by IgE-mediated sensitization. Therefore, the portion of the AD-asthma comorbidity with true mechanistic homology should focus on AA. The present study thus takes the comorbidity of AD and AA as the core subject of investigation. To elucidate the pathophysiological foundation behind the AD and AA comorbidity, the “outside-inside-outside” hypothesis proposed by Peter M Elias and colleagues takes the disruption of the AD skin barrier as the initiating event, triggering immune dysregulation that drives systemic allergic sensitization and extending “from the outside in” to affect the airways, ultimately leading to AA ( 8 ). Although this systemic Th2 activation state conceptually links the allergic inflammatory manifestations of the skin and the respiratory system, the molecular mechanisms that span the two tissues remain unclear. Moreover, previous biomarker studies have mostly been confined to the respective disease backgrounds of AD or AA in isolation, and systematic transcriptomic research targeting the cross-phenotype of “AD-AA comorbidity” can further the understanding of atopic comorbidity and facilitate the search for comorbid biomarkers with diagnostic value. In this study, we performed an integrative, multi-stage transcriptomic analysis to identify and validate a core comorbid gene linking AD and AA. First, using bulk RNA sequencing datasets for AD and AA obtained from the Gene Expression Omnibus (GEO), we combined differential expression analysis (DEA) and weighted gene co-expression network analysis (WGCNA) with machine learning-based feature engineering to screen candidate core comorbid genes, which were then integrated into multiple algorithmic combinations to construct a comorbidity diagnostic model. The model’s performance was assessed on an internal test set and further confirmed in independent external validation cohorts. Second, we analyzed single-cell transcriptomic data from AD and AA using the Scissor, scAB, and DEGAS algorithms to identify disease-associated pathogenic cell subgroups, characterize the expression patterns of the comorbid gene within these subgroups, and elucidate its functional relevance through cell-cell communication analysis and virtual gene knockouts. Third, we used spatial transcriptomic data generated by 10x Visium and Xenium to validate these findings in situ , revealing spatial co-occurrence between the target cell subsets that highly express the comorbid gene and effector cells. Finally, we validated the expression of the comorbid gene in tissue-specific cellular models. The overall workflow of this study is illustrated in Figure 1 . Figure 1 Flow chart of the study. 2 Materials and methods 2.1 Data source and processes Publicly available transcriptomic datasets for AD and AA were obtained from the GEO database. For the discovery stage, bulk RNA sequencing datasets were used, comprising two AD cohorts (GSE121212, GSE130588) and one AA cohort (GSE41861). Three additional AD datasets (GSE120721, GSE16161, GSE36842) served as an internal testing set, while GSE32924 (AD) and GSE19187 (AA) were used as independent external validation cohorts. For single-cell transcriptomic analyses, one AD dataset (GSE222840) and one AA dataset (GSE193816) were included. For spatial transcriptomic validation, a 10x Visium dataset for AD (GSE206391) and a Xenium dataset for AA (GSE269354) were used. Supplementary Table 1 summarizes comprehensive details for all included datasets, including accession numbers, specimen/tissue origins, sample sizes (control/disease cohorts), sequencing platforms, and technologies. 2.2 Differential expression analysis and WGCNA To achieve a sufficient sample size and enhance statistical power, two independent AD datasets from different transcriptomic platforms (GSE121212 based on RNA-seq and GSE130588 based on microarray) were integrated. The RNA-seq data were pre-transformed into continuous log2-CPM values using the cpm function within the ‘edgeR’ package in R (version 4.4.2) to ensure cross-platform scale compatibility. Inherent batch effects and platform leverage between the datasets were subsequently eliminated using the ComBat algorithm via the “sva” R package. Principal Component Analysis (PCA) was conducted before and after integration to confirm the removal of batch effects. For the AA dataset (GSE41861), which was generated on a single microarray platform, raw intensity values were directly log2-transformed and normalized without requiring cross-platform integration. DEA was performed on the batch-corrected, integrated AD expression matrix and the normalized AA expression matrix separately using the “limma” R package to identify genes differentially expressed between the disease and control groups, where for the AD analysis, platform source was additionally incorporated into the linear model as a blocking factor to intercept any residual batch variance. Differentially expressed genes (DEGs) for AD and AA were defined using a threshold of |log 2 fold change (FC)| > 0.585 and adjusted P 30% mitochondrial content, as well as doublet filtration via biologically informed marker exclusions and Scrublet) ( 10 ), was converted from h5ad to Seurat using the “scBridge” R package. Finally, the FindNeighbors and FindClusters functions were applied to identify cell clusters, and cell types were annotated by consulting the CellMarker database and the existing literature ( 10 , 11 ). 2.5 Phenotype-associated pathogenic cell subset identification using Scissor, scAB, and DEGAS To identify cell subsets contributing to disease onset, we used two complementary methods, Scissor (based on regression models and correlation analysis) ( 12 ) and scAB (a knowledge- and graph-guided matrix factorization model) ( 13 ) to integrate clinical phenotypes with bulk RNA-seq and scRNA-seq data. Using the R packages Scissor and scAB to extract clinical phenotypes from AD (GSE121212/GSE130588) and AA (GSE41861), we mapped cell-level clinical correlations onto the corresponding scRNA-seq matrices to identify core pathogenic cell subpopulations Scissor+/scAB+. Additionally, to robustly map patient-level disease phenotypes onto individual cells via multi-trait transfer learning, the deep learning-based framework DEGAS (Diagnostic Evidence GAuge of Single cells) ( 14 ) was implemented using the “DEGAS” R packages (v1.0.0), linked to a Python environment. Highly variable genes from the AD and AA single-cell datasets were intersected with the bulk RNA expression matrices of the AD and AA discovery cohorts, respectively, to obtain common gene sets. The DEGAS preprocess Counts function was used to normalize the matrices, and the disease labels in the bulk RNA data, along with the single-cell cluster cell types, were converted into one-hot encodings. The DEGAS model, featuring a dense neural network, was trained using the runCCMTLBag function, which included a 3-layer network architecture and 5 resampling repetitions. Finally, the predClassBag function predicted the cell-level disease-associated probabilities, and the association scores were mapped back into the single-cell metadata slots to cross-validate the pathogenic cell subpopulations identified by Scissor and scAB. Subsequently, to address the sparsity and high variability of single-cell data, we performed pseudo-bulk analysis on the pathogenic subpopulations using the AggregateExpression function of the “Seurat” R package. Genes with a base mean > 10 and an adjusted P < 0.05 were identified as more robust DEGs. 2.6 Gene regulatory network inference via pySCENIC To explore upstream transcriptional networks governing pathogenic cell-fate decisions, pySCENIC (v0.12.1) was used under Python (v3.7) for gene regulatory network (GRN) inference ( 15 ). Subpopulations from over 500 cells in the AD/AA dataset were downsampled to 500 and compiled into a loom object. First, GRNBoost2 inferred raw transcription factor (TF)-target co-expression modules. Second, cis-regulatory motifs within ±10 kbp of TSS were scanned via the hg38-cisTarget database to prune indirect targets, yielding high-confidence regulons. Third, AUCell scored single-cell regulon activity. Finally, the regulator-specific scores (RSSs) were calculated for pathogenic subgroups of differentially expressed comorbid genes, and rank scatterplots were plotted based on the rankings. Simultaneously, upstream core TFs associated with comorbid genes were identified via reverse-gene-based analysis, and an interaction network diagram was constructed. 2.7 In silico perturbation To evaluate the pathogenic roles of core genes associated with comorbidity and their upstream transcription factors, Geneformer was first employed for in silico overexpression of the core genes, followed by the application of scTenifoldKnk to in silico knock out these genes. Finally, CellOracle was used to computationally perturb the upstream transcription factors. Geneformer is a Transformer-based deep learning model ( 16 ). For network analysis, a 6-layer pre-trained Geneformer foundation transformer model (GF-6L-30M-I2048, v0.0.1) was deployed under Python (v3.10). The model was fine-tuned by retaining the pathogenic and effector cell subpopulations from the AD/AA dataset, with a linearly decaying learning rate of 5e-5, L2 regularization strength of 0.001, 500 warm-up steps, 10 training epochs, and GPU acceleration. The number of genes per cell was fixed at 2048 via adaptive truncation or padding. Then, we calculated cosine similarities between the healthy, overexpression, random, and disease states and det
## Related Clinical Research

- [Contextual factors affecting routine immunisation delivery in Mumbunda Health Zone, DRC — head nur](https://medichelpline.com/clinical-feed/medrxiv-2-contextual-factors-shaping-the-delivery-of-routine-immunisation-services-in-the.md)
- [Kennedy renews attacks on vaccines at Children’s Health Defense conference](https://medichelpline.com/clinical-feed/stat-news-1-kennedy-renews-attacks-on-vaccines-at-children-s-health-defense-conference.md)
- [ASNS-dependent asparagine supports intestinal ILC2 responses and type 2 immunity under nutrient li](https://medichelpline.com/clinical-feed/frontiers-in-immunology-15-asns-dependent-asparagine-availability-supports-intestinal-ilc2-responses-and.md)
- [Allergic airway inflammation alters coronavirus outcomes: protection for SARS-CoV-2 and MERS-CoV,](https://medichelpline.com/clinical-feed/biorxiv-21-allergic-airway-inflammation-protects-against-sars-cov-2-and-mers-cov-disease.md)
- [Respiratory viruses trigger ferroptosis-like lipid peroxidation and mitochondrial remodeling in lu](https://medichelpline.com/clinical-feed/pubmed-42663456.md) (DOI: 10.1128/spectrum.01957-26)

## Navigation
- [← Back to Infectious Disease Feed](https://medichelpline.com/clinical-feed/infectious-disease.md)
- [← All Clinical Specialties](https://medichelpline.com/clinical-feed.md)
## Medical & Regulatory Disclaimer

> [!CAUTION]
> MedicHelpline content is structured for research, educational, and professional discovery purposes. It does not constitute individual medical advice, clinical diagnosis, or treatment recommendations.
> Always verify dosing, contraindications, and regulatory alerts against official product labeling and primary regulatory sources before clinical decision-making.