Explainable machine learning approach for cancer prediction through binarilization of RNA sequencing data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Machine learning models show promise for rapid cancer diagnosis using RNA sequencing data.
• Explaining predictions from complex, high-dimensional RNA sequencing data remains a significant challenge.

Purpose of the Study:

• To propose and evaluate a novel data binarization technique for RNA sequencing data.
• To develop explainable cancer prediction models using this technique.
• To assess the impact of binarization on model performance and feature importance.

Main Methods:

• Applied a binarization technique to RNA sequencing data.
• Constructed five machine learning models: neural network, random forest, xgboost, support vector machine, and decision tree.
• Utilized four cancer datasets from the National Cancer Institute Genomic Data Commons.
• Evaluated model performance using metrics for imbalanced datasets (geometric mean, MCC, F-Measure, AUC).

Main Results:

• The binarization approach yielded comparative performance across models.
• The method required fewer features for prediction.
• Data binarization significantly improved model explainability by clarifying feature contributions.
• The technique demonstrated potential in enhancing both performance and interpretability.

Conclusions:

• Data binarization is a viable technique for processing RNA sequencing data in cancer prediction.
• This method enhances the explainability of machine learning models in oncology.
• The approach offers a promising avenue for developing more transparent and effective diagnostic tools.