Interpretable Multi-Cancer Early Detection Using SHAP-Based Machine Learning on Tumor-Educated Platelet RNA.

Area of Science:

• Biomarkers and Omics
• Computational Biology and Bioinformatics
• Oncology and Cancer Research

Background:

• Tumor-educated platelets (TEPs) are a novel source for non-invasive multi-cancer early detection (MCED).
• Existing machine learning (ML) applications in TEP data lack explainability for gene-level insights.
• Integrating interpretability into ML for TEP analysis is crucial for uncovering biological mechanisms.

Purpose of the Study:

• To develop an interpretable ML framework for cancer detection using platelet RNA sequencing data.
• To combine high predictive performance with biological insights for early cancer diagnosis.
• To identify gene-level contributions and regulatory associations in TEPs for MCED.

Main Methods:

• Analyzed 2018 TEP RNA samples from 18 tumor types using seven ML classifiers.
• Applied SHAP (Shapley Additive Explanations) for model interpretability and feature analysis.
• Utilized GeneMANIA for network analysis to uncover regulatory insights and pathway tracing.

Main Results:

• Neural networks (NN, DNN) and XGBoost achieved high predictive performance (AUC ~0.93), accurately detecting early-stage cancers.
• SHAP analysis identified key biomarkers (e.g., SLC38A2, DHCR7, IFITM3) and conditional gene interactions (e.g., USF3, ARL2, DSTN).
• Identified NFYC as a shared transcriptional hub, revealing context-dependent regulatory patterns.

Conclusions:

• Interpretable ML applied to TEP RNA data provides robust biomarkers for early cancer detection.
• The framework reveals context-dependent regulatory patterns crucial for understanding cancer biology.
• TEPs offer a promising, information-rich, non-invasive medium for early cancer screening.