AI2030 Responsible AI Tools Marketplace

Key Takeaways

Graph-based AI models show promise for complex medical diagnosis but face a critical barrier to real-world use: most lack the interpretability that regulators and clinicians require. The review found that post-hoc explanation methods (like SHAP and Gradient Saliency) are being adapted from other AI domains rather than being purpose-built for graph models, suggesting the field is still developing specialized tools for this application.
Different explanation methods reveal different biological insights with real trade-offs. When tested on Alzheimer's disease data, SHAP and Sensitivity Analysis identified the broadest range of known disease pathways, while Gradient Saliency and Graph Masking uncovered different metabolic signatures—meaning the choice of explanation method fundamentally affects which medical insights clinicians see.
Current AI systems for medical data integration are computationally expensive relative to their transparency gains. The benchmark found that methods offering "deeper biology" (SHAP and Graph Masking) require significantly higher computational costs than faster alternatives, creating a practical challenge for clinical deployment where both accuracy and efficiency matter.
The field is dominated by static, oversimplified models that may miss real-world complexity. Most of the 26 studies reviewed use static graphs built from simple similarity measures, while emerging trends like dynamic topology learning and hierarchical graph structures remain underdeveloped—suggesting current clinical applications may not capture how biological relationships actually change over time.
Cancer diagnosis dominates research, leaving other diseases understudied. The review's focus on Alzheimer's disease for benchmarking highlights a gap: most existing interpretable graph models target cancer, meaning other diseases lack equivalent validation of these explanation methods.

Abstract

Integrating heterogeneous biomedical data including imaging, omics, and clinical records supports accurate diagnosis and personalised care. Graph-based models fuse such non-Euclidean data by capturing spatial and relational structure, yet clinical uptake requires regulator-ready interpretability. We present the first technical survey of interpretable graph based models for multimodal biomedical data, covering 26 studies published between Jan 2019 and Sep 2024. Most target disease classification, notably cancer and rely on static graphs from simple similarity measures, while graph-native explainers are rare; post-hoc methods adapted from non-graph domains such as gradient saliency, and SHAP predominate. We group existing approaches into four interpretability families, outline trends such as graph-in-graph hierarchies, knowledge-graph edges, and dynamic topology learning, and perform a practical benchmark. Using an Alzheimer disease cohort, we compare Sensitivity Analysis, Gradient Saliency, SHAP and Graph Masking. SHAP and Sensitivity Analysis recover the broadest set of known AD pathways and Gene-Ontology terms, whereas Gradient Saliency and Graph Masking surface complementary metabolic and transport signatures. Permutation tests show all four beat random gene sets, but with distinct trade-offs: SHAP and Graph Masking offer deeper biology at higher compute cost, while Gradient Saliency and Sensitivity Analysis are quicker though coarser. We also provide a step-by-step flowchart covering graph construction, explainer choice and resource budgeting to help researchers balance transparency and performance. This review synthesises the state of interpretable graph learning for multimodal medicine, benchmarks leading techniques, and charts future directions, from advanced XAI tools to under-studied diseases, serving as a concise reference for method developers and translational scientists.

41 pages

Key Takeaways

Graph-based AI models show promise for complex medical diagnosis but face a critical barrier to real-world use: most lack the interpretability that regulators and clinicians require. The review found that post-hoc explanation methods (like SHAP and Gradient Saliency) are being adapted from other AI domains rather than being purpose-built for graph models, suggesting the field is still developing specialized tools for this application.
Different explanation methods reveal different biological insights with real trade-offs. When tested on Alzheimer's disease data, SHAP and Sensitivity Analysis identified the broadest range of known disease pathways, while Gradient Saliency and Graph Masking uncovered different metabolic signatures—meaning the choice of explanation method fundamentally affects which medical insights clinicians see.
Current AI systems for medical data integration are computationally expensive relative to their transparency gains. The benchmark found that methods offering "deeper biology" (SHAP and Graph Masking) require significantly higher computational costs than faster alternatives, creating a practical challenge for clinical deployment where both accuracy and efficiency matter.
The field is dominated by static, oversimplified models that may miss real-world complexity. Most of the 26 studies reviewed use static graphs built from simple similarity measures, while emerging trends like dynamic topology learning and hierarchical graph structures remain underdeveloped—suggesting current clinical applications may not capture how biological relationships actually change over time.
Cancer diagnosis dominates research, leaving other diseases understudied. The review's focus on Alzheimer's disease for benchmarking highlights a gap: most existing interpretable graph models target cancer, meaning other diseases lack equivalent validation of these explanation methods.

Abstract

41 pages

Interpretable graph-based models on multimodal biomedical data integration: A technical review and benchmarking

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links

Interpretable graph-based models on multimodal biomedical data integration: A technical review and benchmarking

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links