svMIL2

svMIL2 predicts the pathogenic effects of somatic non-coding structural variants that disrupt three-dimensional genome architecture.

Key Features:

• Focus on Non-Coding Regions: Targets somatic structural variants in non-coding regions that can disrupt Topologically Associated Domains (TADs) and gene-enhancer regulatory interactions.
• Multiple Instance Learning: Implements a multiple instance learning framework that groups SV disruptions into "bags" of instances rather than requiring fixed feature matrices.
• Pathogenicity Prediction: Predicts pathogenicity of TAD boundary-disrupting SVs by assessing their association with gene expression aberrations within the same sample.
• Differential Interaction Analysis: Compares regulatory interaction patterns of somatic pathogenic SVs to non-pathogenic somatic and germline SVs to identify distinct biological impacts.

Scientific Applications:

• Cancer Research: Identifies candidate non-coding oncogenic drivers by predicting pathogenic effects of somatic non-coding SVs across cancer types.
• Clinical Implications: Supports assessment of non-coding SV pathogenicity for more comprehensive genomic profiles relevant to clinical decision-making and personalized medicine.
• Genomic Studies: Elucidates how structural variants alter 3D chromatin interactions and gene regulation via disruption of TAD boundaries.

Methodology:

Applies a multiple instance learning framework that groups SVs into bags, assesses associations between TAD boundary-disrupting SVs and sample-matched gene expression aberrations to predict pathogenicity, and performs differential interaction analysis comparing somatic pathogenic, non-pathogenic somatic, and germline SVs.