Development of an absolute assignment predictor for triple-negative breast cancer subtyping using machine learning
Fadoua Ben Azzouz1, Bertrand Michel2, Hamza Lasla1
1Unité de Bioinfomique, Institut de Cancérologie de L'Ouest, Bd Jacques Monod, 44805, Saint Herblain Cedex, France; SIRIC ILIAD, Nantes, Angers, France.
Computers in Biology and Medicine
|December 14, 2020
Summary
This study developed a machine learning model to accurately predict triple-negative breast cancer (TNBC) subtypes using a limited set of gene expression indicators, overcoming limitations of previous methods for precision medicine.
Area of Science:
- Oncology
- Genomics
- Bioinformatics
Background:
- Triple-negative breast cancer (TNBC) heterogeneity complicates precision medicine.
- Existing transcriptomics-based TNBC subtyping methods lack sensitivity and are dataset-dependent.
- A robust, universally applicable TNBC subtyping method is needed for routine diagnostics.
Purpose of the Study:
- To develop a highly accurate, dataset-independent machine learning model for TNBC subtyping.
- To identify a minimal set of gene expression probes for reliable intra-patient TNBC diagnosis.
- To overcome batch effects and improve diagnostic test feasibility.
Main Methods:
- Utilized probe binary comparison to create patient-specific indicators.
- Employed combined filter and wrapper methods with cross-validation for optimal probe selection.
- Trained and evaluated random forest, gradient boosting (GB), and extreme gradient boosting models.
- Used nested cross-validation to select the best-performing predictive model.
Main Results:
- Identified fifty key indicators that highlight biological characteristics of TNBC subtypes.
- The gradient boosting (GB) model, using the selected indicators, demonstrated superior performance over other models.
- The developed approach showed potential for robust intra-patient diagnosis, independent of specific datasets.
Conclusions:
- A machine learning model based on a limited set of gene expression indicators can accurately predict TNBC subtypes.
- Gradient boosting offers a promising approach for developing reliable TNBC subtyping diagnostic tools.
- This method addresses limitations of current transcriptomics-based approaches, paving the way for routine clinical application.
