Precision oncology: a review to assess interpretability in several explainable methods

Area of Science:

• Biomedical Informatics
• Computational Biology
• Machine Learning in Healthcare

Background:

• Precision medicine aims to tailor treatments using patient data, but machine learning models, especially deep learning, often lack interpretability.
• Understanding and trusting model decisions is crucial for clinical implementation, beyond just prediction accuracy.

Approach:

• This review compares six machine learning methods (tree-, regression-, kernel-based, and two novel explainable methods) for precision medicine.
• Evaluation focused on accuracy, multi-omics capability, explainability, and implementability for clinical use.
• Gene expression data showed improved performance over mutational status as input for these methods.

Key Points:

• No significant accuracy differences were found between the compared machine learning methods.
• Tree-based methods demonstrated superior interpretability for clinicians compared to other tested algorithms.
• Model comprehension and ease of use are critical challenges in applying machine learning to precision medicine.

Conclusions:

• Machine learning interpretability is key for advancing precision medicine treatments.
• Tree-based algorithms are recommended for their balance of accuracy and interpretability in clinical settings.
• Future research should focus on enhancing the explainability and implementability of complex machine learning models.