Biomarker Discovery in Rare Malignancies: Development of a miRNA Signature for RDEB-cSCC

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Machine learning (ML) aids tumor biomarker identification but struggles with rare cancers due to limited patient data.
• Recessive dystrophic epidermolysis bullosa (RDEB) patients face a high risk of aggressive cutaneous squamous cell carcinomas (cSCCs), necessitating early detection.
• MicroRNAs (miRNAs) show promise as liquid biopsy markers for non-invasive cancer detection.

Purpose of the Study:

• To characterize miRNA signatures in RDEB for potential tumor detection.
• To overcome RDEB sample scarcity by analyzing miRNA profile similarities with other tumor types using TCGA data.
• To develop and validate ML-based predictive models for RDEB-cSCC detection.

Main Methods:

• Analyzed miRNA expression profiles from RDEB primary cells and compared them to TCGA tumor data.
• Trained elastic net logistic regression models using 33, 10, and 3 miRNAs based on head and neck squamous cell carcinoma (HN-SCC) data.
• Validated model performance on independent HN-SCC datasets, RDEB cell-based miRNA-Seq data, and RDEB exosomes.

Main Results:

• Identified significant miRNA similarities between RDEB-SCC and HN-SCC, enabling model training.
• Achieved high predictive performance (AUC-ROC 83-100%) across datasets, including RDEB exosomes, demonstrating clinical potential.
• Developed three distinct ML models with varying miRNA complexity, all showing robust diagnostic capabilities.

Conclusions:

• A diagnostic miRNA signature for RDEB-cSCC detection is feasible using ML and publicly available data.
• The developed models show high accuracy in predicting tumors from cell-based data and exosomes.
• This approach can improve early detection of RDEB-associated cSCCs, facilitating timely intervention.