Multi-View Multiattention Graph Learning With Stack Deep Matrix Factorization for circRNA-Drug Sensitivity...

Area of Science:

• Computational Biology
• Genomics
• Pharmacology

Background:

• Identifying circular RNA (circRNA)-drug sensitivity associations (CDsA) is vital for drug development.
• Traditional wet experiments are inefficient and costly for determining CDsA.
• Existing computational methods have limitations, particularly with false-negative associations.

Purpose of the Study:

• To present a multi-view framework, MAGSDMF, for identifying latent CDsA.
• To overcome limitations in current CDsA prediction methods, especially false negatives.
• To enhance the accuracy and reliability of circRNA-drug association predictions.

Main Methods:

• MAGSDMF employs multiple attention mechanisms and graph learning for feature extraction across circRNA and drug similarity networks.
• Self-Deep Matrix Factorization (SDMF) is used to extract features directly from CDsAs.
• A multi-attention channel mechanism integrates features for CDsA reconstruction, followed by another DMF for latent CDsA identification.
• Contrastive learning (CL) is integrated to improve model generalization.

Main Results:

• MAGSDMF demonstrated superior performance in identifying latent CDsAs on two datasets.
• Achieved high AUC values of 0.9743 and 0.9739 using 5-fold cross-validation.
• Case studies validated the reliability of MAGSDMF in identifying true associations.

Conclusions:

• MAGSDMF provides an effective computational approach for predicting circRNA-drug sensitivity associations.
• The multi-view framework enhances feature extraction and integration for improved prediction accuracy.
• This method holds promise for accelerating drug development by efficiently identifying potential therapeutic targets.