Protocol for obtaining cancer type and subtype predictions using subSCOPE.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate cancer subtyping is crucial for personalized treatment and understanding tumor heterogeneity.
• Existing methods may not fully leverage the potential of multi-omics data for precise classification.

Purpose of the Study:

• To present a standardized protocol for cancer type and subtype prediction using the subSCOPE machine learning method.
• To enable classification of diverse cancer samples using multiple omics data types.

Main Methods:

• The protocol details data preparation, subSCOPE setup, and inference execution.
• It integrates five omics data types: DNA methylation, gene expression, microRNA expression, point mutations, and copy-number variants.
• Supports individual selection of cancer types and data modalities.

Main Results:

• The subSCOPE protocol facilitates subtype-level classification for non-TCGA cancer samples.
• It achieves classification across 26 The Cancer Genome Atlas (TCGA) cancer cohorts and 106 distinct subtypes.
• Demonstrates a flexible framework for leveraging multi-omics data in cancer research.

Conclusions:

• The presented protocol offers a robust and adaptable tool for cancer subtyping using machine learning and multi-omics data.
• It enhances the ability to classify cancer subtypes, particularly for samples outside of large public datasets like TCGA.
• Facilitates deeper insights into cancer biology and supports the development of targeted therapies.