From:  Explainable AI for multi-omics in precision medicine: a systematic review

 Representative applications in cancer, prognosis, and drug response.

Application domainRepresentative studiesMethodsOmics layersXAI contributionClinical relevanceReferences
Cancer subtype classificationAMMO; DeepMoIC; MoAGNNAttention-fusion; GCNs; hierarchical GNNGenomics, transcriptomics, clinicalAttention weights identify key genes and pathways driving subtypesGuides treatment selection; identifies targetable subtypes[63, 86, 100]
Biomarker discoveryDeepKEGG; SHAP-based analysesPathway-informed DL; explainable GNNsMulti-omics (varies)Feature attribution prioritizes genes, pathways for validationNovel therapeutic targets; diagnostic signatures[53, 71, 72]
Drug response predictionDeepFusionCDR; DeepInsight-3DTransformers; CNNs; DLGenomics, transcriptomics, proteomicsSHAP, attention identify resistance/sensitivity mechanismsTreatment selection; avoidance of ineffective therapies[90, 91, 101]
Survival/PrognosisAutosurv; CoFormerSurv; PathformerInterpretable DL; collaborative transformers; pathway-informedMulti-omics + clinicalTime-dependent feature importance; pathway-level explanationsRisk stratification; treatment intensity decisions[55, 56, 84]
Immunotherapy responseImmunotherapy survival XAI; Blood-cancer multi-omicsExplainable ML; SHAPMulti-omics, immune signaturesIdentifies immune-related gene signatures and TMB contributionsPatient selection for checkpoint inhibitors[70, 89]
Single-cell integrationFactVAEFactorized VAEs; multimodal DLSingle-cell multi-omicsDisentangled latent factors; cell-type specific explanationsCellular heterogeneity; developmental trajectories[34, 102]

XAI: explainable artificial intelligence; GCNs: graph convolutional networks; GNNs: graph neural networks; SHAP: SHapley Additive exPlanations; DL: deep learning; ML: machine learning; VAEs: variational autoencoders.