Multi-omics integration method based on attention deep learning network for biomedical data classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Integrating multi-omics data is crucial for understanding human diseases but presents significant bioinformatics challenges.
• Deep learning (DL) methods show promise for multi-omics analysis, yet often overlook patient-specific and cross-omic correlations.

Purpose of the Study:

• To develop a novel deep learning approach for comprehensive multi-omics data integration and biomedical data classification.
• To address limitations in existing DL methods by incorporating patient-omics correlations.

Main Methods:

• Proposed Multi-omics Attention Deep Learning Network (MOADLN) for biomedical data classification.
• Utilized self-attention mechanisms for dimensionality reduction and patient correlation construction within each omics type.
• Employed a Multi-Omics Correlation Discovery Network (MOCDN) to learn cross-omic correlations in the label space.

Main Results:

• MOADLN demonstrated superior performance compared to state-of-the-art methods on datasets including mRNA, DNA methylation, and miRNA expression.
• Identified essential disease biomarkers, showcasing the method's clinical relevance.
• Validated the generalizability of MOADLN on KIRP and KIRC datasets.

Conclusions:

• MOADLN effectively integrates multi-omics data by considering intra-omics patient correlations and cross-omics label-space correlations.
• The proposed deep learning network offers a powerful tool for biomedical data classification and analysis.