Random Forest Modelling of High-Dimensional Mixed-Type Data for Breast Cancer Classification

Area of Science:

• Computational biology
• Bioinformatics
• Cancer research

Background:

• High-throughput technologies generate vast multiomics data for complex diseases like breast cancer.
• Investigating disease aetiologies requires integrative analyses of diverse data types to capture comprehensive biological information.

Purpose of the Study:

• To develop a permutation-based framework for unbiased integration of mixed-type data in random forest methods.
• To enable simultaneous assessment of relative feature importance for multiomics datasets.
• To identify clinically relevant patterns and prognostic biomarkers in breast cancer.

Main Methods:

• A novel permutation-based framework utilizing random forest methods.
• Integration of mixed-type data (e.g., genomics, proteomics) for comprehensive analysis.
• Evaluation through simulation studies, machine learning datasets, and two independent breast cancer cohorts.

Main Results:

• The framework demonstrated minimal multicollinearity and limited overfitting.
• Reproducibility and robustness were confirmed by consistent feature importance and clustering profiles across cohorts.
• An identified cluster showed prognostic value for clinical outcome post-chemotherapy, outperforming existing classifications.

Conclusions:

• The proposed framework facilitates unbiased integration of multiomics data for complex disease research.
• It provides a robust method for identifying prognostic subgroups in breast cancer.
• This approach enhances our understanding of breast cancer heterogeneity and clinical outcome prediction.