DMAGCL: A dual-masked adaptive graph contrastive learning framework for predicting circRNA-drug sensitivity

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Circular RNAs (circRNAs) are crucial regulators of gene expression and drug response.
• Experimental identification of circRNA-drug sensitivity is time-consuming and labor-intensive.

Purpose of the Study:

• To develop an efficient computational framework for predicting circRNA-drug sensitivity.
• To establish a novel method for identifying circRNA-drug associations and informing precision therapy.

Main Methods:

• Introduced DMAGCL, a Dual-Masked Graph Contrastive Learning framework.
• Employed a synergistic dual-masking strategy (path- and edge-level) for robust representation learning.
• Utilized an adaptive contrastive loss with a scheduled temperature parameter and an attention-based fusion classifier (AFC) for cross-modal interactions.

Main Results:

• DMAGCL achieved state-of-the-art performance with average AUC of 0.8940 and AUPR of 0.9006 (five-fold CV).
• Demonstrated superior performance over existing methods like GATECDA and MNGACDA.
• Case studies showed an 80% experimental validation rate for four anticancer drugs, highlighting predictive reliability.

Conclusions:

• DMAGCL establishes a new paradigm for circRNA-drug association prediction through its innovative components.
• The framework provides a robust, interpretable, and scalable computational tool for discovering circRNA-drug associations.
• Offers valuable insights for drug resistance mechanisms and precision therapy design.