Joint global and local interpretation method for CIN status classification in breast cancer

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Breast cancer is a leading cause of cancer diagnoses globally.
• MicroRNAs (miRNAs) are implicated in chromosomal instability (CIN) in various cancers.
• Accurate prediction of CIN status is crucial for improving breast cancer patient outcomes.

Purpose of the Study:

• To develop and validate a novel computational method for predicting CIN status in breast cancer using miRNA expression data.
• To enhance the interpretability of predictive models for breast cancer CIN.
• To identify key miRNAs associated with CIN in breast cancer.

Main Methods:

• A joint global and local interpretation method, GL_XGBoost, was developed.
• GL_XGBoost integrates eXtreme Gradient Boosting (XGBoost) for prediction and SHapley Additive exPlanations (SHAP) for feature selection and interpretation.
• The model was validated using the TCGA-BRCA dataset.

Main Results:

• GL_XGBoost achieved an accuracy of 78.57% and an Area Under the Curve (AUC) of 0.87 in predicting CIN status.
• SHAP analysis effectively identified miRNA features strongly correlated with CIN.
• Comprehensive visualizations provided global and local model interpretability.

Conclusions:

• GL_XGBoost offers an effective and interpretable approach for predicting breast cancer CIN from miRNA data.
• The method aids in understanding the microscopic relationship between miRNAs and CIN.
• This approach has the potential to improve breast cancer patient stratification and survival prediction.