Advancing sarcoma diagnostics with expanded DNA methylation-based classification.

Area of Science:

• Oncology
• Genomics
• Computational Biology

Background:

• Sarcomas present significant diagnostic challenges, requiring differentiation from other mesenchymal tumors for appropriate clinical management.
• Previous work established a machine learning-based sarcoma classifier using DNA methylation data.
• This study introduces a major advancement in the scale and precision of this classifier.

Purpose of the Study:

• To present a significantly advanced sarcoma classifier with increased scale and precision.
• To improve the diagnostic accuracy and confidence for mesenchymal tumors.
• To facilitate the clinical implementation of a molecular diagnostic tool for sarcomas.

Main Methods:

• Analysis of DNA methylation profiles and histologic data from a large, multi-institutional cohort of mesenchymal tumors.
• Development of an updated machine learning classifier (v13.1) trained on 4377 methylation profiles across 116 tumor sub-classes and control groups.
• Rigorous validation using five-fold nested cross-validation and four independent cohorts (1547 tumors).

Main Results:

• The sarcoma classifier v13.1 achieved 98% class-level accuracy and a Brier score of 0.017.
• Predictions were obtained in 73% of validation cases, with 91% matching original histopathology diagnoses.
• The classifier provided definitive molecular diagnoses or reclassifications in 6% of cases with ambiguous histology.

Conclusions:

• The enhanced sarcoma classifier demonstrates increased diagnostic predictions and improved concordance with histologic evaluation.
• This advancement is expected to promote the clinical adoption of the tool for diagnosing mesenchymal tumors.
• The updated classifier represents a significant step forward in the molecular diagnosis of sarcomas.