New Approach for Risk Estimation Algorithms of BRCA1/2 Negativeness Detection with Modelling Supervised Machine...

Area of Science:

• Oncology
• Bioinformatics
• Machine Learning

Background:

• BRCA1/2 gene testing is crucial for breast cancer treatment selection and secondary cancer risk assessment.
• Current BRCA1/2 testing is expensive, time-consuming, and labor-intensive.
• Accurate identification of BRCA1/2 mutation status is vital for personalized oncology.

Purpose of the Study:

• To develop a machine learning algorithm for predicting BRCA1/2 negativity in high-risk breast cancer patients.
• To identify key clinical, demographic, and genetic features predictive of BRCA1/2 negativity.
• To reduce the need for direct BRCA1/2 gene testing, saving time and resources.

Main Methods:

• A comprehensive dataset of 125 features from 2070 high-risk breast cancer patients was compiled over 20 years.
• Clinical, demographic, and genetic data were numeralized and normalized for machine learning analysis.
• k-Nearest Neighbors (KNN) and Decision Tree (DT) models were employed to detect BRCA1/2 negativity.

Main Results:

• An algorithm utilizing 9 key features achieved high accuracy in predicting BRCA1/2 negativity.
• Feature reduction and optimization increased algorithm accuracy compared to standard DT models.
• The developed algorithm identified BRCA1/2 negativity with 92.88% accuracy in minutes.

Conclusions:

• The algorithm enables rapid and accurate prediction of BRCA1/2 negativity in high-risk breast cancer patients.
• This approach mitigates the stress, time, and cost associated with traditional BRCA1/2 gene testing.
• The algorithm can significantly expedite clinical practice and treatment planning for breast cancer patients.