Interpretable multi-instance heterogeneous graph network learning modelling CircRNA-drug sensitivity association...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (circRNAs) influence drug sensitivity in human cells, impacting therapeutic outcomes.
• Traditional experimental methods for identifying circRNA-drug sensitivity associations are inefficient and costly.
• Accurate prediction of novel circRNA-drug sensitivity relationships is essential for advancing personalized medicine.

Purpose of the Study:

• To develop an effective computational method for predicting circRNA-drug sensitivity associations.
• To construct a heterogeneous graph network model integrating circRNA and drug features.
• To enhance the accuracy and interpretability of circRNA-drug sensitivity predictions.

Main Methods:

• Constructed a heterogeneous network using circRNA features, drug features, and drug structural information.
• Employed a heterogeneous graph convolutional network (GCN) for deep feature embedding.
• Integrated multi-instance learning (MIL) with a pseudo-metapath instance generator and BiTrans for metapath-level representation.
• Developed an interpretable multiscale attention network for final prediction and analysis.

Main Results:

• The MiGNN2CDS model demonstrated superior prediction accuracy compared to existing state-of-the-art methods.
• Case studies confirmed the model's capability in predicting previously unknown circRNA-drug associations.
• Model interpretability was validated through high-confidence meta-path analysis.

Conclusions:

• MiGNN2CDS offers a powerful and interpretable approach for predicting circRNA-drug sensitivity.
• The findings contribute to the development of targeted therapies by identifying key circRNA biomarkers.
• The study provides a valuable computational tool for drug sensitivity research, with code and data publicly available.