Machine learning unlocks new insights in ovarian cancer proteomics

Proteomics has revolutionized cancer research, offering profound insights into the molecular underpinnings of oncogenesis. In this study, we reanalyze an ovarian cancer proteomics dataset using advanced machine learning (ML) techniques to uncover novel biomarkers and therapeutic targets. The dataset, comprising 189 proteomic profiles from the PRIDE database, includes samples from serous ovarian cancer, non-cancerous tissues, and plasma. Data were split into 80% for training and 20% for validation.

Our approach integrates unsupervised clustering, supervised classification, and feature selection algorithms. Supervised ML models, including deep learning networks, achieve an impressive 94% accuracy in predicting and diagnosing ovarian cancer, significantly outperforming traditional methods.

Feature selection techniques identify a ranked list of 50 key proteins, used for the further pathway analysis. In addition to standard ovarian cancer biomarkers presented in the OVA1 diagnostic test (CA125, transthyretin (prealbumin), apolipprotein A1 (Apo A-1), beta-2 microglobulin, and transferrin) the list of proteins exhibiting specific expression patterns includes von Willebrand factor, actin, cytoplasmic 1, attractin and immunoglobulin lambda-like polypeptide 1.
This reanalysis not only provides a comprehensive proteomic profile of ovarian cancer but also underscores the potential of deep learning models for its early prediction and diagnosis. Our findings demonstrate the power of ML in transforming proteomics data into actionable biological insights, paving the way for improved preclinical cancer screening and personalized management of ovarian cancer.

In summary, this work represents a significant step forward in the application of ML to cancer proteomics, demonstrating the potential to unlock new dimensions of understanding in ovarian cancer research.