<![CDATA[Artificial intelligence increasingly supports genomic diagnostics, yet complex biological data and opaque predictive mechanisms limit clinical trust and biological interpretation. This study aimed to develop an explainable artificial intelligence framework that improves diagnostic accuracy while revealing the genomic features underlying model decisions. A retrospective quantitative computational design was applied to 2,480 biological samples containing genomic, molecular, and clinical information. Logistic regression, support vector machine, random forest, extreme gradient boosting, deep neural network, and an explainable artificial intelligence model were trained and evaluated using stratified training, validation, and independent testing subsets. Diagnostic performance was assessed through accuracy, sensitivity, specificity, F1-score, area under the receiver operating characteristic curve, calibration, and explanation stability. Biological plausibility was examined through recurrent feature rankings and pathway-level interpretation procedures. The proposed model achieved the best performance, with 0.923 accuracy, 0.931 sensitivity, 0.916 specificity, 0.925 F1-score, and 0.968 AUC. Stable feature-attribution patterns were associated with higher classification reliability, while unstable explanations were more frequent in borderline and misclassified cases. The findings indicate that explainability can strengthen genomic diagnostics by combining predictive performance with interpretable biological evidence. The framework offers a rigorous basis for trustworthy genomic decision support, although external, multi-center, and prospective validation remains necessary before clinical implementation.]]>
Copyrights © 2026
