Subtype-MGTP: a cancer subtype identification framework based on multi-omics translation.

Area of Science:

• Computational biology
• Bioinformatics
• Cancer research

Background:

• Cancer subtyping is crucial for research and treatment, with multi-omics data integration offering an efficient strategy.
• Current methods often rely on genomics data, but protein expression data provides a closer phenotypic representation.
• Integrating scarce protein data with genomics data for cancer subtyping presents significant challenges, including data scarcity and balancing omics-specific and cross-omics learning.

Purpose of the Study:

• To develop a novel cancer subtyping framework, Subtype-MGTP, that effectively integrates multi-omics data, particularly addressing the challenge of limited protein expression data.
• To enhance cancer subtyping accuracy by leveraging predicted protein expression data derived from genomics data.
• To improve the integration of omics-specific and cross-omics learning in multi-omics data analysis.

Main Methods:

• Subtype-MGTP framework utilizes a translation module to predict protein expression from multi-type genomics data, guided by available protein data.
• An improved deep subspace clustering module with contrastive learning is employed to cluster the predicted protein expression data for refined subtyping.
• The framework is evaluated on benchmark datasets against nine state-of-the-art cancer subtyping methods.

Main Results:

• Subtype-MGTP significantly outperforms nine existing state-of-the-art cancer subtyping methods on benchmark datasets.
• The identified cancer subtypes demonstrate interpretability through clinical and survival analysis.
• The framework shows robustness to missing protein data and excels in integrating imbalanced multi-omics datasets.

Conclusions:

• Subtype-MGTP offers a powerful and robust approach for cancer subtyping by effectively integrating multi-omics data, particularly when protein data is limited.
• The novel translation and deep subspace clustering modules enhance subtyping accuracy and provide interpretable results.
• This framework advances multi-omics data integration strategies for cancer research and personalized medicine.