Machine Learning Framework for Ovarian Cancer Diagnostics Using Plasma Lipidomics and Metabolomics

Area of Science:

• Gynecologic Oncology
• Metabolomics
• Machine Learning

Background:

• Ovarian cancer (OC) is a leading gynecologic malignancy with distinct metabolic alterations.
• Liquid biopsy offers potential for early OC detection through metabolic profiling.

Purpose of the Study:

• To develop and validate a machine learning pipeline for OC detection using plasma metabolomics data.
• To identify optimal biomarker combinations and machine learning models for distinguishing OC from benign and control samples.

Main Methods:

• Integrated lipidomics (HPLC-MS) and NMR-based metabolomics on plasma samples from 229 subjects (103 OC patients, 107 benign, 19 controls).
• Systematic evaluation of feature selection methods (Mann-Whitney, Kruskal-Wallis, SVM-RFE) and machine learning architectures (CNN, XGBoost).

Main Results:

• A CNN model achieved 81% accuracy in distinguishing OC from benign cases.
• XGBoost with SVM-RFE achieved 96% accuracy differentiating benign from control samples.
• Multiclass classification yielded up to 78% accuracy using one-versus-one CNN models.

Conclusions:

• Machine learning, particularly deep learning and ensemble methods, shows significant potential for tailored OC diagnostic applications.
• The study provides a methodological framework for multi-omics biomarker discovery in gynecologic oncology.
• Findings offer biological insights into OC pathophysiology, supporting integrated approaches.