SLInterpreter: An Exploratory and Iterative Human-AI Collaborative System for GNN-Based Synthetic Lethal Prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence in Oncology

Background:

• Synthetic lethal (SL) relationships offer targeted cancer therapy potential but require robust discovery methods.
• Current AI models lack sufficient interpretability and alignment with domain expertise, hindering experimental validation.
• High experimental costs necessitate improved AI-driven approaches for identifying SL interactions.

Purpose of the Study:

• To develop an iterative Human-AI collaborative framework for refining synthetic lethal (SL) relationship discovery.
• To enhance the interpretability and biological relevance of AI-driven predictions for targeted cancer therapies.
• To facilitate mechanism analysis and uncover novel SL relationships through expert-AI synergy.

Main Methods:

• Human-Engaged Knowledge Graph Refinement using metapath strategies for domain knowledge integration.
• Cross-Granularity SL Interpretation Enhancement and Mechanism Analysis for comparing predictions and uncovering biological insights.
• Iterative cycles of AI model optimization and expert intervention to build trust and improve accuracy.

Main Results:

• The framework successfully refines knowledge graphs and enhances SL interpretation across different granularities.
• It aids experts in organizing and comparing AI predictions, leading to improved understanding of potential SL relationships.
• The iterative collaboration fosters trust and ensures that AI-generated insights align with biological principles.

Conclusions:

• The proposed Human-AI framework significantly improves the discovery and interpretation of synthetic lethal relationships for cancer therapy.
• Iterative collaboration enhances AI model performance and ensures alignment with domain-specific biological knowledge.
• This approach offers a promising solution for overcoming the limitations of current AI models in precision oncology.