CancerSubtyper: a deep learning framework for cancer subtyping through DNA methylation data

Area of Science:

• Computational Biology
• Genomics
• Oncology

Background:

• Molecular subtyping is crucial for precision oncology, classifying tumors into relevant categories.
• DNA methylation is a promising biomarker for cancer subtyping.
• Current DNA methylation analysis faces challenges like high dimensionality, batch effects, and lack of user-friendly tools.

Purpose of the Study:

• To present CancerSubtyper, an automated computational framework for deep learning-based cancer subtyping using DNA methylation data.
• To provide an intuitive web interface for interactive exploration and downstream analysis of cancer subtypes.
• To lower the barrier for large-scale methylation analysis in precision oncology research.

Main Methods:

• Developed an end-to-end computational framework integrating semi-supervised and hybrid learning models.
• Implemented automated preprocessing, feature selection, and batch correction.
• Integrated interactive visualization tools for subtype exploration and validation.

Main Results:

• CancerSubtyper provides a deep learning-based approach for cancer subtyping using DNA methylation.
• The framework integrates two complementary models for classifying known subtypes and identifying novel ones.
• An intuitive web interface facilitates interactive exploration and downstream analysis.

Conclusions:

• CancerSubtyper offers an automated, end-to-end workflow for molecular subtyping via DNA methylation.
• The user-friendly web interface makes large-scale methylation analysis more accessible.
• This tool empowers precision oncology research by facilitating robust cancer subtyping.