HCMAF: Hierarchical Feature Aggregation and Cross-Modal Attention Fusion Framework for Multi-Omics Patient...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Large-scale multi-omics datasets offer potential for precise disease treatment.
• Integrating multi-omics data is challenging due to complex inter- and intra-omics relationships.

Purpose of the Study:

• To propose a novel framework, Hierarchical Feature Aggregation and Cross-Modal Attention Fusion (HCMAF), for effective multi-omics data integration.
• To enhance patient classification and biomarker identification using integrated multi-omics data.

Main Methods:

• HCMAF framework incorporates three modules: Hierarchical Feature Aggregation Graph Attention (HGAT) for intra-omics features, Cross-Modal Attention (CMA) for inter-omics complementarity, and Confidence-driven Multi-omics Fusion (CMF) for dynamic integration.
• HGAT captures topological features via neighborhood aggregation.
• CMA models cross-omics dependencies.
• CMF uses learnable weights for integrating omics-specific predictions.

Main Results:

• HCMAF demonstrated superior patient classification performance across four benchmark datasets.
• The proposed framework consistently outperformed existing multi-omics integration methods.
• Component analysis validated the essential contributions of HGAT, CMA, and CMF modules.

Conclusions:

• HCMAF provides an effective approach for precise multi-omics data integration.
• The framework enhances capabilities in patient classification and biomarker discovery.
• The developed model is publicly available in Python with Pytorch.