Comparison of XAI methods with advantages and limitations.
| Method category | Specific methods | Core idea | Advantages | Limitations | Use in multi-omics | References |
|---|---|---|---|---|---|---|
| Model-agnostic (Post-hoc) | SHAP | Shapley values for feature attribution | Theoretically grounded; consistent; global and local explanations | Computationally expensive; approximations may be biased | Identifying key genes, pathways; biomarker prioritization | [45, 64] |
| LIME | Local surrogate models | Fast; model-agnostic; intuitive | Unstable; local approximations may not generalize | Patient-specific explanations; clinical decision support | [65, 66] | |
| Permutation importance | Performance drop after feature shuffling | Simple; efficient; global ranking | Ignores interactions; assumes feature independence | Comparing omics layer importance; feature screening | [26, 28] | |
| Partial dependence | Marginal effect visualization | Reveals directionality; intuitive plots | Independence assumption; unrealistic extrapolation | Understanding feature-outcome relationships | [27, 28] | |
| Model-specific | Attention weights | Learned importance scores from attention | Intrinsic interpretability; captures interactions; hierarchical | Faithfulness concerns; may not reflect true importance | Pathway-level explanations; cross-modal fusion | [67] |
| Saliency maps | Gradients of output w.r.t. inputs | Computationally efficient; fine-grained | Noisy; gradient saturation; unstable | Genomic sequence interpretation; imaging-omics fusion | [68, 69] | |
| Sparse/Rule-based | LASSO, decision trees, rule lists | Intrinsically interpretable; faithful | May sacrifice performance; limited non-linear interactions | Feature selection; interpretable final-layer predictors | [22, 54, 58] |
SHAP: SHapley Additive exPlanations; XAI: explainable artificial intelligence; LIME: Local Interpretable Model-agnostic Explanations; w.r.t.: with respect to; LASSO: Least absolute shrinkage and selection operator.
During the preparation of this work, the authors used Google AI tools to improve language and readability. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.
MH: Conceptualization, Writing—original draft, Visualization. SH: Writing—review & editing, Supervision. KA: Methodology, Writing—review & editing, Formal analysis. MW: Methodology, Formal analysis, Writing—review & editing. All authors have read and approved the final version of the manuscript.
The authors declare that they have no conflicts of interest.
Not applicable.
Not applicable.
Not applicable.
No new data were generated or analyzed in this review. All referenced datasets are publicly available from the sources cited.
Not applicable.
© The Author(s) 2026.
Open Exploration maintains a neutral stance on jurisdictional claims in published institutional affiliations and maps. All opinions expressed in this article are the personal views of the author(s) and do not represent the stance of the editorial team or the publisher.