Using a machine learning approach to identify key prognostic molecules for esophageal squamous cell carcinoma

Area of Science:

• Oncology
• Bioinformatics
• Molecular Biology

Background:

• Esophageal squamous cell carcinoma (ESCC) exhibits high molecular heterogeneity, challenging the reproducibility of existing prognostic biomarkers.
• Identifying reliable biomarkers is crucial for improving patient outcomes and treatment strategies in ESCC.

Purpose of the Study:

• To identify optimal prognostic biomarkers for ESCC utilizing machine learning algorithms.
• To overcome the limitations of low reproducibility in previously reported ESCC biomarkers.

Main Methods:

• Literature search identified 48 biomarkers linked to ESCC recurrence or prognosis.
• A molecular interaction network was constructed, and functional modules were identified.
• Machine learning algorithms (LR, SVM, ANN, RF, XGBoost) were employed for prognostic classification and feature selection using TCGA and GEO datasets.
• Prognostic classifier performance was evaluated using AUC, and molecule importance was ranked by frequency.

Main Results:

• A molecular network with 3 functional modules and 17 component molecules was established.
• Stratifin (SFN) was identified as the optimal prognostic biomarker for ESCC based on its high occurrence frequency in machine learning classifiers.
• The prognostic value of SFN was validated in two independent cohorts.

Conclusions:

• Stratifin (SFN) is a highly reproducible and optimal prognostic biomarker for esophageal squamous cell carcinoma.
• The study highlights the utility of machine learning in identifying robust biomarkers from complex molecular data.