MoAGL-SA: a multi-omics adaptive integration method with graph learning and self attention for cancer subtype...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Deep learning enhances multi-omics data integration for cancer subtype classification.
• Challenges exist in embedding sample structure and flexible integration strategies.

Purpose of the Study:

• To develop an adaptive multi-omics integration method for improved cancer subtype classification.
• To address limitations in feature space embedding and integration flexibility.

Main Methods:

• Proposes MoAGL-SA, an adaptive multi-omics integration method using graph learning and self-attention.
• Generates patient relationship graphs and extracts omic-specific embeddings using graph convolutional networks.
• Employs self-attention for adaptive weighting of omics data for integration.

Main Results:

• MoAGL-SA outperforms existing algorithms on breast, kidney renal papillary cell, and kidney renal clear cell carcinoma datasets.
• Successfully identifies key biomarkers for breast invasive carcinoma.
• Demonstrates improved feature learning and multi-omics data integration.

Conclusions:

• MoAGL-SA offers a robust approach for cancer subtype classification.
• The method effectively integrates diverse omics data by adaptively weighting features.
• Identified biomarkers can aid in understanding cancer biology and developing targeted therapies.