Explainable machine learning reveals the role of taste-related genetic variants in body mass index: a nutrigenetic perspective.

BACKGROUND: Taste perception-related genetic variants may influence dietary behavior and energy balance; however, their relationship with body mass index (BMI) remains unclear. This study aimed to investigate the association between taste-related genetic variants and BMI using both statistical analysis and an explainable machine learning approach. METHODS: This retrospective observational study included 200 individuals aged 5-60 years with available genotype and BMI data. Variants in TAS2R38, TAS1R2, TAS1R3, and FGF21 were analyzed. Associations with BMI categories were evaluated using chi-square tests. A Random Forest classification model was developed, and SHAP (SHapley Additive exPlanations) values were used to quantify the contribution of each variant and explore allele-dose effects. RESULTS: No significant associations were found between genetic variants and BMI categories in adults (p > 0.05). In children, rs35744813 showed a significant inverse association with BMI (p = 0.011). SHAP analysis revealed that TAS2R38 variants (rs10246939 and rs1726866) and FGF21 rs838133 demonstrated the highest relative contributions to BMI classification. Allele-dose patterns suggested differences in model attribution across genotype groups. CONCLUSION: Taste-related genetic variants did not demonstrate independent associations with BMI classification in the present study. However, explainable machine learning approaches identified heterogeneous model contributions across genetic variants, suggesting potential differences in genotype-related patterns within a multifactorial framework. These findings may contribute to future research exploring gene-phenotype relationships and precision nutrition approaches.