BRCA1-specific machine learning model predicts variant pathogenicity with high accuracy
Mohannad Khandakji1,2, Hind Hassan Ahmed Habish3, Nawal Bakheet Salem Abdulla3
1Division of Genomics and Translational Biomedicine, College of Health and Life Sciences, Hamad Bin Khalifa University, Doha, Qatar.
Physiological Genomics
|June 19, 2023
Summary
A new machine learning model accurately predicts the pathogenicity of BRCA1 variants, aiding in breast cancer risk assessment. This tool identified potentially harmful BRCA2 variants in Qatari patients for further study.
Area of Science:
- Genomics
- Computational Biology
- Oncology
Background:
- Clinical annotation of novel BRCA1 variants lags behind identification, necessitating advanced computational tools for accurate risk assessment.
- Breast cancer risk is significantly influenced by germline mutations in BRCA1 and BRCA2 genes.
- Variants of Uncertain Significance (VUS) pose challenges in clinical genetic testing and patient management.
Purpose of the Study:
- To develop a specialized machine learning model for predicting the pathogenicity of all BRCA1 variants.
- To apply the developed BRCA1 model and a prior BRCA2 model to assess VUS in Qatari breast cancer patients.
- To enhance the clinical interpretation of BRCA variants and improve breast cancer risk stratification.
Main Methods:
- An XGBoost machine learning model was developed using variant features (position, frequency, consequence) and in silico prediction scores.
- The model was trained and validated using BRCA1 variants classified by the Evidence-Based Network for the Interpretation of Germline Mutant Alleles (ENIGMA) consortium.
- Performance was further assessed on an independent set of missense VUS with experimentally determined functional scores.
Main Results:
- The BRCA1 model achieved high accuracy (99.9%) in predicting pathogenicity for ENIGMA-classified variants.
- The model demonstrated strong performance (93.4% accuracy) in predicting functional consequences for an independent set of missense VUS.
- The models identified 2,115 potentially pathogenic BRCA1 variants from the BRCA Exchange database and four potentially pathogenic BRCA2 variants in Qatari patients.
Conclusions:
- The developed BRCA1 machine learning model is highly effective for predicting variant pathogenicity, addressing a critical gap in clinical annotation.
- Application of BRCA-specific models to Qatari breast cancer patients revealed no pathogenic BRCA1 variants but highlighted potentially pathogenic BRCA2 variants for validation.
- These computational tools offer significant potential for improving genetic risk assessment and guiding clinical management of breast cancer patients.
