netDx: interpretable patient classification using integrated patient similarity networks.

Area of Science:

• Biomedical informatics
• Computational biology
• Machine learning in healthcare

Background:

• Accurate patient classification is crucial for diagnosis, prognosis, and treatment.
• Genomic data integration and interpretability are key challenges in clinical prediction.
• Existing machine learning models often lack generalizability and interpretability.

Purpose of the Study:

• To introduce netDx, a novel supervised patient classification framework.
• To demonstrate netDx's accuracy, generalizability, and interpretability in clinical prediction.
• To showcase netDx's utility in identifying predictive biological pathways.

Main Methods:

• Developed netDx, a framework based on patient similarity networks.
• Applied netDx to a cancer survival benchmark dataset with multiple data types.
• Compared netDx performance against traditional machine learning approaches.
• Utilized pathway-level gene expression for defining patient similarity.

Main Results:

• netDx significantly outperformed most machine learning approaches in cancer survival prediction across multiple cancer types.
• netDx provided more interpretable results by visualizing decision boundaries in patient similarity space.
• Pathway-level gene expression analysis with netDx identified key biological pathways associated with patient outcomes in breast cancer and asthma.

Conclusions:

• netDx offers a powerful and interpretable tool for patient classification and clinical prediction.
• The framework facilitates the discovery of biological features relevant to disease prognosis.
• netDx serves as both a predictive classifier and a discovery engine for biomedical research.