AI-Derived Biomarkers in Endometriosis

Decoding Endometriosis: Biomarkers and Immune Landscape Identified by Machine Learning

Key Points

Highlights: 

• Machine learning–based analysis identifies candidate gene biomarkers associated with endometriosis.
• Distinct alterations in the immune microenvironment of eutopic endometrium are observed.
• Integration of transcriptomics and AI enables development of a diagnostic prediction model.

Importance:

• Non-invasive, biology-based diagnostic approaches remain a major unmet need in endometriosis, particularly for early and subtle disease.

What's Done Here?

• This is a bioinformatics study integrating RNA-seq datasets from multiple publicly available genomic datasets (GEO cohorts).
• Differentially expressed genes were identified and refined using WGCNA, LASSO, SVM, and random forest algorithms.
• Six candidate biomarkers (DDX56, TAL1, ALX3, DDX6, ADRBK2, ZMYND11) were selected.
• An artificial neural network model was developed and externally validated.
• Immune cell composition in eutopic endometrium was analyzed usinga computational approach, CIBERSORT.

Key Results:

• Six genes demonstrated potential diagnostic value, with differential expression in endometriosis versus controls.
• The AI-based model showed high diagnostic performance in both training and validation datasets.\
• Altered immune cell composition was observed, including increased CD8⁺ T cells, follicular helper T cells, and monocytes.
• Decreased proportions of resting dendritic cells, macrophages, and mast cells were also identified.
• Findings support a role for immune dysregulation in the pathophysiology of endometriosis.

Strengths and Limitations:

• Strengths are: integration of multiple transcriptomic datasets; use of complementary machine learning approaches; external validation of the diagnostic model; combined analysis of gene expression and immune landscape.
• Limitations are: retrospective bioinformatic design; reliance on public datasets; limited clinical validation; small number of samples for immunohistochemistry; uncertain generalizability to broader patient populations.

From the Editor-in-Chief – EndoNews

"The application of machine learning to endometriosis research reflects a broader shift toward data-driven discovery in complex diseases. By integrating transcriptomic datasets and applying multiple analytical algorithms, this study identifies candidate biomarkers and proposes a diagnostic model while simultaneously exploring the immune microenvironment of the eutopic endometrium.

The strength of the work lies in this layered approach. Rather than relying on a single analytical method, the authors combine feature selection techniques with model development and external validation, aiming to improve robustness. The parallel evaluation of immune cell composition further aligns the findings with current understanding that endometriosis involves not only ectopic lesions but also systemic and local immunologic alterations.

At the same time, the distinction between statistical performance and biological validity remains critical. High diagnostic accuracy within curated datasets does not necessarily translate into clinical utility. Publicly available transcriptomic cohorts are inherently heterogeneous, often limited in clinical annotation, and subject to selection bias. As such, model performance may reflect dataset characteristics as much as underlying disease biology.

The identification of candidate genes is similarly interpretative. While differential expression and algorithmic selection highlight potential relevance, these markers are not yet validated as disease drivers, nor are they established as reliable diagnostic tools across diverse patient populations. The immune findings, although consistent with the concept of dysregulation, are inferred from computational deconvolution rather than direct measurement and therefore require cautious interpretation.

Importantly, this study underscores a recurring challenge in endometriosis research: the gap between molecular insight and clinical application. The disease is heterogeneous in presentation, phenotype, and symptom burden, and any diagnostic model must ultimately account for this variability. Without prospective validation in well-characterized cohorts, the translational impact of such models remains uncertain.

This work contributes meaningfully by organizing complex data into a coherent analytical framework and by reinforcing the role of immune–molecular interactions in endometriosis. The next step is not further algorithmic refinement alone, but rigorous biological validation and clinical testing to determine whether these findings can inform diagnosis or patient management in real-world settings."

Lay Summary

Diagnosing endometriosis remains a major challenge, often requiring invasive procedures and long delays before confirmation.

A study published in Frontiers in Immunology by Zhang et al.  from China explores whether gene expression patterns and immune cell changes in the uterine lining could help improve diagnosis by biomarkers.

Using existing genomic data from multiple patient groups, the researchers applied advanced computational methods to identify a small set of genes that differ between women with and without endometriosis. Based on these findings, they developed a model that was able to distinguish between the two groups with high accuracy in both initial and independent datasets.

In addition to gene-level changes, the study also identified differences in the immune environment of the endometrium. Certain immune cells were more abundant, while others were reduced, suggesting that immune imbalance may play a role in the disease process.

While these results provide insight into the biological mechanisms of endometriosis and suggest a potential direction for non-invasive diagnostics, the findings are based on retrospective data and require further validation in clinical settings.

Overall, the study highlights how combining genomic data with advanced analytical approaches may contribute to a more precise understanding of endometriosis, although translation into clinical practice remains to be established.

Research Source: https://pubmed.ncbi.nlm.nih.gov/41830823/