PathHDNN: a pathway hierarchical-informed deep neural network framework for predicting immunotherapy response and...

Area of Science:

• Computational biology
• Cancer research
• Immunology

Background:

• Immunotherapy offers significant survival benefits for cancer patients, but predicting response remains challenging.
• Understanding the mechanisms of immunotherapy efficacy is crucial for improving patient outcomes.

Purpose of the Study:

• To develop a novel deep learning framework, PathHDNN, for predicting immunotherapy responses.
• To identify key biological pathways associated with immunotherapy efficacy.

Main Methods:

• Developed a pathway hierarchical-informed deep neural network (PathHDNN) framework.
• Encoded biological pathway hierarchy, somatic mutations, and copy number variations into the network.
• Optimized network edge weights while preserving biological interpretability.

Main Results:

• PathHDNN outperformed existing machine learning methods and biomarkers in predicting immunotherapy response across cohorts.
• Identified critical pathways, including p53-mediated tumor suppression and JAK2-mediated immune regulation, in melanoma.
• Demonstrated PathHDNN's ability to provide interpretable insights into treatment response mechanisms.

Conclusions:

• PathHDNN offers an effective framework for predicting patient responses to cancer immunotherapy.
• The model provides valuable insights into the biological mechanisms of treatment response, advancing precision medicine.