The MOGA multi-modal framework based on graph augmentation networks for drug response prediction.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Personalized cancer treatment is hindered by tumor heterogeneity.
• Current computational methods struggle with effective multi-omics data integration for drug response prediction.

Purpose of the Study:

• To present MOGA, an advanced computational framework designed for accurate drug response prediction.
• To address limitations in multi-omics integration for precision oncology.

Main Methods:

• Constructed a heterogeneous graph integrating cell line multi-omics, drug chemical structures, and response data.
• Utilized a relational graph convolutional network (RGCN) to model complex interactions.
• Implemented a type-specific graph augmentation strategy to preserve sensitive signals and reduce noise.

Main Results:

• MOGA significantly outperformed existing computational baselines in predicting drug response (AUROC and AUPR).
• Ablation studies and case analyses confirmed the importance of multi-omics integration and the augmentation mechanism.

Conclusions:

• MOGA demonstrates significant potential for advancing precision oncology through improved drug response prediction.
• The proposed graph augmentation strategy is key to MOGA's enhanced performance.