A random forest-based predictive model for classifying BRCA1 missense variants: a novel approach for evaluating the...

Area of Science:

• Genomics and Bioinformatics
• Cancer Genetics
• Machine Learning in Medicine

Background:

• Accurate classification of BRCA1 variants is crucial for early detection and prevention of breast and ovarian cancers.
• Existing predictive tools often lack interpretability and struggle with specific variant classifications.
• There is a need for high-performance, interpretable tools for clinical significance assessment of BRCA1 variants.

Purpose of the Study:

• To develop an accurate and interpretable predictive tool for classifying the clinical significance of BRCA1 variants.
• To improve upon the limitations of current variant classification methods.
• To enhance pre-symptomatic disease detection and preventive strategies for hereditary cancers.

Main Methods:

• Collected a dataset of BRCA1 benign and pathogenic missense variants.
• Prepared the dataset by analyzing variant effects on protein sequence, incorporating physicochemical changes and conservation scores.
• Trained a random forest-based machine learning model, BRCA1-Forest, for variant classification.

Main Results:

• BRCA1-Forest demonstrated superior performance compared to SIFT, PolyPhen2, CADD, and DANN across multiple evaluation metrics (precision, FPR, AUC ROC, AUC-PR, MCC).
• The model achieved high specificity and sensitivity in classifying BRCA1 variant significance.
• Outperformed existing methods in all tested metrics except for recall.

Conclusions:

• BRCA1-Forest offers a significant advancement in the accurate and interpretable classification of BRCA1 variants.
• The tool has the potential to improve clinical decision-making for patients at risk of hereditary breast and ovarian cancers.
• The developed software is publicly available for research and clinical application.