M&M: an RNA-seq based pan-cancer classifier for paediatric tumours

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• Diagnosing rare pediatric tumors is challenging, historically relying on histology and morphology.
• Advances in genome-wide profiling, like RNA sequencing, enable molecular classification tools.

Purpose of the Study:

• To develop a machine learning algorithm for pan-paediatric cancer classification, specifically including rare tumor types.
• To create a single, comprehensive diagnostic tool for diverse pediatric cancer samples.

Main Methods:

• Developed M&M, an ensemble-based machine learning algorithm utilizing RNA sequencing data.
• Tailored the algorithm for a pan-paediatric cancer approach, accommodating rare tumor types.

Main Results:

• M&M accurately classifies 52 tumor types (99% precision, 80% recall) and 96 subtypes (96% precision, 70% recall).
• Validation on internal and external datasets showed high precision (98-99%) and recall (76-77%).
• Performance is comparable to existing RNA sequencing or methylation-based classifiers.

Conclusions:

• M&M's pan-cancer design facilitates clinical implementation with a single classifier for all diagnostic samples.
• The algorithm's performance rivals specialized classifiers, offering potential to enhance diagnostic accuracy for pediatric cancers.
• Improved diagnostic accuracy can lead to better patient survival and quality of life.