Information-incorporated sparse hierarchical cancer heterogeneity analysis

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Cancer heterogeneity analysis is crucial for precision medicine but often overlooks data sparsity and integrates limited data types.
• Existing methods typically analyze single data types, failing to capture complex hierarchical structures present in combined pathological imaging and omics data.
• Identifying subgroup-specific important features is challenging due to sparsity, a critical aspect often ignored in current analyses.

Purpose of the Study:

• To develop a novel sparse hierarchical heterogeneity analysis framework.
• To integrate multimodal data, specifically pathological imaging and omics data, for a more comprehensive analysis.
• To leverage prior knowledge from literature to enhance feature selection in cancer heterogeneity analysis.

Main Methods:

• Proposed a novel sparse hierarchical heterogeneity analysis framework.
• Integrated pathological imaging and omics data to capture hierarchical structures.
• Incorporated prior knowledge from scientific literature to guide feature selection.

Main Results:

• The developed framework demonstrated satisfactory statistical properties.
• The approach showed competitive numerical performance in simulations.
• A real-world data analysis on TCGA data confirmed the practical value of the approach.

Conclusions:

• The proposed framework effectively analyzes data heterogeneity and sparsity by integrating multimodal data and prior knowledge.
• This approach offers a significant advancement over existing methods by addressing limitations in single-data type analysis and feature selection.
• The study highlights the potential of the framework for improving precision medicine through more robust cancer subtyping.