An Efficient and Interpretable Machine Learning Model for Classifying Breast Cancer Subtypes Using Gene Expression Profiles

Breast Cancer (BRCA) is a complex and heterogeneous disease. This heterogeneity has been shown to affect gene expression patterns and molecular activity of different subtypes in different ways. BRCA subtype identification is of critical importance in the context of prognosis and treatment decisions for the disease. Advances in transcriptomic profiling and Machine Learning (ML) models have enabled the classification of BRCA subtypes with higher accuracy, yet the majority of classification models lack interpretability, thereby limiting their clinical applicability. In this study, an interpretable ML framework for classifying BRCA subtypes is proposed using high-dimensional RNA-sequencing data. The framework was evaluated using a publicly available TCGA transcriptomic dataset, by applying dimensionality reduction techniques and optimizing ML models through grid search tuning. Shapley Additive Explanations (SHAP) values are used to find important transcriptomic markers that facilitate the classification of subtypes. This approach provides insights into the gene sets associated with the molecular mechanisms of each subtype. The experimental results demonstrate that the proposed method exhibits superior performance in terms of accuracy, precision, F1-score, and interpretability when compared to existing works. Finally, the gene set enrichment analysis highlights key pathways associated with BRCA and its subtypes.