DeepCMS: A Feature Selection-Driven Model for Cancer Molecular Subtyping with a Case Study on Testicular Germ Cell...

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Cancer is a complex, heterogeneous disease with molecular variations leading to distinct subtypes.
• Molecular subtyping aids in understanding cancer variability and developing personalized medicines.
• High-dimensional 'omics' data presents challenges like data scarcity and feature dimensionality in subtyping.

Purpose of the Study:

• To propose DeepCMS, a novel deep learning framework for cancer molecular subtyping.
• To leverage feed-forward neural networks, gene set enrichment analysis, and feature selection.
• To construct a representative feature subset for improved subtyping accuracy.

Main Methods:

• Gene expression data transformed into enrichment scores (>22,000 features).
• Selection of top 2000 features for deep learning application.
• Utilized colon cancer gene expression data for framework development and validation.

Main Results:

• DeepCMS demonstrated superior performance over state-of-the-art models.
• Achieved aggregated accuracy, sensitivity, specificity, and balanced accuracy exceeding 0.90 on independent datasets.
• Framework showed generalizability and robustness, applicable to diverse gene expression data.

Conclusions:

• DeepCMS enables accurate and robust classification of cancer molecular subtypes.
• The framework utilizes a compact and informative feature set for enhanced precision.
• This approach holds promise for advancing precision oncology applications.