Interpretable Dynamic Directed Graph Convolutional Network for Multi-Relational Prediction of Missense Mutation and...

Area of Science:

• Oncology
• Computational Biology
• Bioinformatics

Background:

• Tumor heterogeneity complicates drug response prediction, as missense mutations within a single gene can yield diverse outcomes.
• Existing deep learning models for drug response prediction often lack interpretability, functioning as
• black boxes
• and failing to capture intricate mutation-drug relationships.

Purpose of the Study:

• To develop an advanced analytical framework for predicting drug responses in the context of tumor heterogeneity.
• To enhance the interpretability of deep learning models in oncology drug response prediction.

Main Methods:

• Proposed an Interpretable Dynamic Directed Graph Convolutional Network (IDDGCN) framework.
• Utilized directed graphs to distinguish sensitivity and resistance, dynamically updated node weights, and explored intra-gene mutation associations.
• Integrated a weighted mechanism and ground truth construction for enhanced model interpretability and transparency.

Main Results:

• The IDDGCN framework demonstrated superior predictive performance compared to existing state-of-the-art models.
• Both qualitative and quantitative evaluations confirmed the model's strong interpretability, providing transparent explanations for its predictions.
• The model effectively captures complex relationships between missense mutations and drug response.

Conclusions:

• The IDDGCN offers a powerful and interpretable approach for predicting cancer drug responses, addressing limitations of current methods.
• This framework provides a novel perspective for precision oncology, aiding in targeted drug development and personalized treatment strategies.