Related Experiment Videos
Interpretable graph-based models on multimodal biomedical data integration: a technical review and benchmarking
Alireza Sadeghi1, Farshid Hajati2, Ahmadreza Argha3,4
1Holcombe Department of Electrical and Computer Engineering, Clemson University, Clemson, SC, USA.
Nature Communications
|June 16, 2026
Summary
Interpretable graph models enhance multimodal biomedical data analysis for disease classification. Benchmarking explainable AI (XAI) methods reveals complementary strengths for trustworthy AI in healthcare.
Area of Science:
- Biomedical Informatics
- Artificial Intelligence
- Data Science
Background:
- Integrating diverse biomedical data is crucial for healthcare insights.
- Graph-based models excel at capturing complex relationships in data.
- Clinical adoption of these models requires high interpretability.
Purpose of the Study:
- To review interpretable graph-based models for multimodal biomedical data.
- To highlight trends in disease classification, graph construction, and explainability.
- To benchmark explainable artificial intelligence (XAI) techniques.
Main Methods:
- Survey of interpretable graph-based models in biomedical research.
- Categorization of XAI techniques for graph analysis.
- Benchmarking of SHAP, saliency, sensitivity, and graph masking on Alzheimer's disease data.
Main Results:
- Identified dominant trends in disease classification and static graph construction.
- Demonstrated complementary strengths of various XAI techniques.
- Provided a benchmark for explainability methods on real-world biomedical data.
Conclusions:
- Interpretable graph models are vital for trustworthy biomedical AI.
- XAI methods offer valuable insights into model predictions.
- Future work should explore dynamic graphs and LLM-based explainability.
