Advanced machine learning framework for enhancing breast cancer diagnostics through transcriptomic profiling.

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in healthcare

Background:

• Accurate breast cancer diagnostics are crucial for effective treatment.
• Integrating transcriptomic data offers a powerful approach for molecular profiling.
• Machine learning models can enhance diagnostic accuracy and interpretability.

Purpose of the Study:

• To develop an advanced machine learning (ML) framework for breast cancer diagnostics.
• To integrate transcriptomic profiling with optimized feature selection and classification.
• To improve diagnostic accuracy and model interpretability in breast cancer detection.

Main Methods:

• Analysis of 1759 samples (987 breast cancer, 772 controls) using feature selection (Recursive Feature Elimination, Boruta, ElasticNet).
• Dimensionality reduction via Non-Negative Matrix Factorization (NMF), Autoencoders, and transformer embeddings (BioBERT, DNABERT).
• Training and evaluation of classifiers (XGBoost, LightGBM, ensemble voting) with cross-validation and external validation on 175 samples.

Main Results:

• XGBoost and LightGBM achieved high test accuracies (0.91, 0.90) and AUC (0.92), especially with NMF and BioBERT.
• Ensemble Voting demonstrated superior external validation accuracy (0.92), indicating robustness.
• Transformer embeddings and advanced feature selection outperformed conventional methods like PCA.

Conclusions:

• The proposed ML framework significantly enhances breast cancer diagnostic accuracy and interpretability.
• The model demonstrates strong generalizability on an independent dataset.
• Findings support the framework's potential for precision oncology and personalized diagnostics.