Pathogenicity Prediction of Gene Fusion in Structural Variations: A Knowledge Graph-Infused Explainable Artificial...

Area of Science:

• Genomics
• Bioinformatics
• Artificial Intelligence

Background:

• Cancer genome analysis reveals numerous structural variants (SVs) beyond single nucleotide variants (SNVs).
• Identifying driver variants, especially those involving gene fusions, is challenging for clinical practice.
• Accurate artificial intelligence (AI) predictions are crucial for selecting driver variants and improving diagnostic reliability.

Purpose of the Study:

• To develop an explainable AI (XAI) model for predicting the pathogenicity of SVs involving gene fusions.
• To enhance AI-driven genomic analysis for more reliable cancer diagnoses.
• To adapt existing XAI technology for SVs with gene fusions by expanding the knowledge graph and refining the algorithm.

Main Methods:

• Development of an explainable AI (XAI) based on prior work for SNV pathogenicity prediction.
• Augmentation of the knowledge graph with new data specific to SVs and gene fusions.
• Improvement of the AI algorithm to handle complex gene fusion variants.
• Validation of prediction accuracy against existing tools.

Main Results:

• The developed XAI achieved prediction accuracy comparable to existing tools for SVs with gene fusions.
• The XAI successfully provided explanations for its predictions, detailing the reasoning behind variant pathogenicity.
• Explanations were demonstrated to be plausible concerning fundamental pathogenic mechanisms using variant examples.

Conclusions:

• The XAI offers a promising approach for identifying driver SVs with gene fusions in cancer.
• The explainability feature enhances the reliability of AI-driven diagnoses in genomic medicine.
• This work represents a significant advancement toward AI-supported clinical decision-making in cancer genomics.