CanID: A Robust and Accurate RNA-seq Expression-based Diagnostic Classification Scheme for Pediatric Malignancies.

Area of Science:

• Computational biology
• Genomics
• Machine learning in oncology

Background:

• Accurate cancer subtype classification is crucial for personalized medicine.
• Existing methods face challenges in pediatric cancer due to evolving standards and analytical limitations.

Purpose of the Study:

• To develop a robust machine learning classification scheme for pediatric cancer subtypes using transcriptomic data.
• To address the limitations of current classifiers in precision and scope for pediatric cancers.

Main Methods:

• Developed CanID, a stacked ensemble machine learning model.
• Utilized gene-level RNA sequencing count data as the sole input.
• Trained on 3203 pediatric cancer samples across 13 solid tumor and 38 hematologic malignancy subtypes.

Main Results:

• Achieved 99% accuracy for solid tumors and 92%-93% for hematologic malignancies on external datasets.
• Demonstrated robustness against data collection variations, class imbalance, and potential mislabeling.
• Successfully classified challenging subtypes often difficult for clinical histology.

Conclusions:

• CanID offers a highly accurate and robust transcriptome-based approach for pediatric cancer diagnosis and stratification.
• This method advances tumor diagnosis and supports clinically meaningful stratification.
• The CanID tool is publicly available on GitHub for broader research application.