Detecting optimal biomarkers in ovarian cancer cells from high-dimensional mRNA expression data using machine...

Area of Science:

• Computational Oncology
• Genomics
• Biomarker Discovery

Background:

• High-dimensional transcriptomic data presents challenges in identifying reliable biomarkers for ovarian cancer.
• Traditional methods often face overfitting and poor generalization due to data complexity and limited sample sizes.

Purpose of the Study:

• To identify an optimal, biologically meaningful subset of messenger RNA (mRNA) biomarkers for distinguishing ovarian cancer from healthy samples.
• To develop and validate an integrated machine learning-based feature selection framework for biomarker discovery.

Main Methods:

• Analysis of mRNA expression data from ovarian cancer and control cell lines (~63,000 transcripts).
• Implementation of a hybrid feature selection pipeline (statistical filtering, recursive elimination, regularization) with stratified cross-validation.
• Validation using Logistic Regression, Random Forest, XGBoost, Support Vector Machine classifiers, and droplet digital PCR (ddPCR).

Main Results:

• Identification of 80 discriminative mRNA biomarkers with perfect classification performance (accuracy, sensitivity, specificity = 1.00).
• Experimental validation via ddPCR confirmed expression patterns, including downregulation of ADAMTS12, FN1, ABI3BP and overexpression of EPCAM, COX6C, TMT1B.
• Pathway enrichment analysis indicated involvement in DNA repair, RNA processing, protein translation, immune regulation, and metabolic reprogramming.

Conclusions:

• The hybrid feature selection framework effectively reduces dimensionality and enhances biomarker reliability in ovarian cancer research.
• Identified mRNA signatures are biologically interpretable and show potential for diagnostic and therapeutic applications in ovarian cancer.