OncoTrace-TOO: Interpretable Machine Learning Framework for Cancer Tissue-of-Origin Identification Using...

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Cancer of unknown primary (CUP) presents a significant diagnostic challenge, limiting targeted therapy options.
• Current machine learning and transcriptomic methods for tumor origin identification often lack interpretability and struggle with similar tumor types.

Purpose of the Study:

• To develop a transparent and biologically interpretable machine learning framework for accurate cancer tissue-of-origin (TOO) classification.
• To facilitate clinical diagnosis and improve treatment strategies for CUP.

Main Methods:

• Developed OncoTrace-TOO, a novel tissue-of-origin classification model utilizing gene expression profiles.
• Employed pan-cancer discriminative molecular features identified via one-vs-rest differential expression analysis.
• Utilized logistic regression as the classification algorithm.

Main Results:

• OncoTrace-TOO achieved an overall accuracy of 0.967, with perfect classification for seven cancer types.
• Demonstrated high predictive accuracy on TCGA and GEO validation datasets for both primary and metastatic cancers.
• Showcased enhanced ability to resolve histologically similar malignancies and classify rare subtypes, with 0.857 accuracy on independent clinical samples.

Conclusions:

• OncoTrace-TOO provides high predictive accuracy for tissue-of-origin classification and biologically meaningful insights.
• The framework supports clinical decision-making, promising improved diagnostic precision for challenging cancer cases.
• Offers potential for guiding personalized treatment strategies in oncology.