Integrative clinical–genomic modeling evaluates exploratory Non-APOE candidate signals in Alzheimer’s disease

Alzheimer’s disease (AD) is characterized by a genetic architecture dominated by APOE , while predictive modeling in clinically enriched cohorts is often driven by neuropsychological and clinical variables. This combination limits the interpretability of secondary genetic signals in multimodal frameworks. We applied a two-stage Korean WES–WGS design across the Alzheimer’s continuum. Candidate loci were nominated in a whole-exome sequencing (WES) amnestic mild cognitive impairment (aMCI) cohort ( n = 346) and evaluated in an independent whole-genome sequencing (WGS) cohort ( n = 995) with epidemiologic and neuropsychological variables. Gradient-boosted decision trees (XGBoost) were evaluated using a common stratified five-fold cross-validation scheme with pooled out-of-fold (OOF) predictions. Preprocessing and one-hot encoding were fitted within each training fold only. Identifier/linkage fields were excluded before modeling, and a redundant categorical age-band variable was removed. Feature attribution and APOE sensitivity analyses were used for descriptive model interpretation. The full clinical–genomic model showed near-perfect internal discrimination (AUROC 0.998; 95% CI 0.996–0.999) but included diagnostic-proximal clinical and neuropsychological variables and was therefore interpreted as an upper-bound clinical separability model rather than evidence of prospective predictive utility. In the strict leakage-reduced sensitivity analysis, AUROC decreased to 0.713 (95% CI 0.682–0.745) and decreased further to 0.625 (95% CI 0.589–0.658) after removal of rs429358 and APOE proxy information. The five-SNP model had modest discrimination (AUROC 0.637; 95% CI 0.605–0.670), whereas the Non-APOE SNP-only model was weak (AUROC 0.533; 95% CI 0.497–0.568). In fold-wise validation permutation analysis of the strict model, Test_Age ranked first overall and rs429358 was the highest-ranking genetic feature; the Non-APOE candidates had small or near-zero mean AUROC decreases after permutation. This study presents a Korean WES–WGS exploratory clinical–genomic candidate-evaluation framework. The findings indicate that high multimodal discrimination in clinically enriched AD datasets can reflect diagnostic-proximal clinical separability, while Non-APOE candidate variants should be interpreted as hypothesis-generating model features rather than confirmed susceptibility loci, independent predictors, or causal variants. A second independent Korean cohort with harmonized candidate genotypes and comparable clinical and neuropsychological variables was not available within the present study, precluding direct external replication, and the WES and WGS stages used different clinical endpoints.