A novel machine learning algorithm selects proteome signature to specifically identify cancer exosomes

Area of Science:

• Biochemistry
• Oncology
• Bioinformatics

Background:

• Non-invasive cancer diagnosis faces challenges with low sensitivity and specificity.
• Exosomes, containing parent cell biomarkers, are abundant in biological fluids.
• A flexible exosome-based method for pan-cancer diagnosis is needed.

Purpose of the Study:

• To develop a machine learning (ML) model for distinguishing cancers using exosome protein biomarkers.
• To identify universal exosome protein biomarkers for cancer detection.
• To classify cancer subtypes and differentiate cancer exosomes from others.

Main Methods:

• Utilized datasets of exosome proteins from various human samples (cell lines, tissue, plasma, serum, urine).
• Identified key exosome proteins: Clathrin Heavy Chain (CLTC), Ezrin (EZR), Talin-1 (TLN1), Adenylyl cyclase-associated protein 1 (CAP1), and Moesin (MSN).
• Developed random forest ML models to analyze protein panels for cancer detection and subtyping.

Main Results:

• Achieved Area Under the Receiver Operating Characteristic Curve (AUROC) scores > 0.91 for models using plasma, serum, or urine exosome proteins.
• Demonstrated superior performance compared to Support Vector Machine, K Nearest Neighbor, and Gaussian Naive Bayes classifiers.
• Successfully identified protein panels that distinguish cancer exosomes and aid in classifying cancer subtypes.

Conclusions:

• Established a reliable protein biomarker signature for cancer exosomes.
• Validated the ML approach for sensitive and specific non-invasive cancer diagnosis.
• The method offers scalable ML capability for improved early cancer detection.