AI-Driven Variant Annotation for Precision Oncology in Breast Cancer.

Area of Science:

• Genomic Variant Interpretation
• Precision Oncology
• Computational Biology

Background:

• Interpreting genomic variants in breast cancer is challenging, especially for rare variants lacking clinical guidance.
• Current methods focus on common mutations, leaving many variants of unknown significance unclassified.

Purpose of the Study:

• To develop an AI/ML framework for systematic identification of breast cancer variants linked to key phenotypes like ESR1 and EZH2 activity.
• To integrate diverse datasets (genomic, transcriptomic, structural, drug response) for a structure-informed variant classification approach.

Main Methods:

• Utilized CCLE/DepMap and TCGA datasets to analyze over 12,000 breast cancer variants.
• Employed an AI/ML framework integrating multi-omics data to identify structurally clustered mutations with shared functional consequences.
• Shifted classification from frequency-based to structure-informed analysis.

Main Results:

• Identified novel associations between PIK3CA, TP53, and other gene variants with ESR1 signaling, impacting endocrine therapy response predictions.
• Discovered EZH2-associated variants in unexpected genomic contexts, suggesting potential epigenetic therapy targets.
• Expanded the repertoire of potentially actionable mutations through structure-informed classification.

Conclusions:

• The AI/ML framework offers a scalable, clinically relevant method to accelerate variant annotation in breast cancer.
• Findings provide new insights into drug sensitivity and resistance, enabling improved patient stratification and drug repurposing.
• This approach redefines cancer variant interpretation, bridging genomics, functional biology, and precision medicine for personalized treatments.