vERnet-B

vERnet-B predicts the pathogenicity of missense single-nucleotide variants in the BRCT domain of BRCA1 by applying a deep convolutional neural network to features derived from AlphaFold2-predicted tertiary protein structures.

Key Features:

• Target scope: Focuses on missense single-nucleotide variants within the BRCT domain of BRCA1.
• AI-Based model: Employs a deep convolutional neural network to recognize variant pathogenicity.
• Structural integration: Learns pathogenicity-associated features from AlphaFold2-predicted tertiary protein structures rather than relying solely on primary amino acid sequences.
• Performance and validation: Reported accuracy of 85% for identifying disease-associated BRCA1 variants, with balanced false-positive and true-positive rates and the ability to identify the pathogenicity of variant A1708E where AlphaFold2 alone was insufficient.

Scientific Applications:

• Clinical interpretation of VUS: Assists interpretation of variants of uncertain significance in BRCA1 for clinical genetics contexts.
• Structural-functional variant analysis: Provides insights into phenotypic consequences of missense variants by leveraging tertiary structure-derived features.
• Support for genetic counseling and risk assessment: Supplies pathogenicity predictions that can inform personalized medicine decisions and risk assessment.

Methodology:

Uses AlphaFold2-predicted tertiary protein structures as input features and applies a deep convolutional neural network to learn pathogenicity-associated three-dimensional structural features of BRCA1 BRCT-domain missense single-nucleotide variants.