Mapping the Analytical Landscape of Gene-Diet Interactions in Epidemiology: From Classical Models to Causal and...

Diet is a major, modifiable determinant of cardiometabolic, cancer, and inflammatory disease risk, yet individuals frequently exhibit substantial heterogeneity in metabolic and clinical responses to similar dietary exposures. Genetic susceptibility and its interplay with diet plausibly contribute to this variability, motivating gene-diet (G×D) interaction research and the broader ambition of precision nutrition. Translation has lagged, however, because interaction effects are typically modest, context-dependent, and difficult to reproduce, particularly in the presence of pervasive dietary measurement error, heterogeneous exposure definitions, and stringent multiplicity correction. A methodologically oriented synthesis is presented across eight domains of contemporary G×D epidemiology: classical regression interaction models; efficient study designs; dietary assessment and measurement error; dietary patterns, mixtures, and non-linear modeling; genome-wide and polygenic approaches; causal inference frameworks; multi-omics integration; and machine learning. Central concepts include the recognition that "interaction" is a scale-dependent estimand and that transparent reporting of coding choices and effect-modification metrics-including additive interaction when relevant for public health interpretation-is essential. Credible inference further depends on high-quality, harmonized dietary phenotyping with explicit energy adjustment and, where feasible, biomarker calibration, alongside robust control of population structure and gene-diet correlation using ancestry adjustment, mixed models, and family-based designs. Genome-wide and polygenic risk-based approaches expand discovery potential but require disciplined multiplicity strategies, discovery-replication workflows, and explicit evaluation of portability and equity across ancestries. Causal inference methods can strengthen etiologic interpretation when assumptions are defensible and sensitivity analyses are routinely implemented. Multi-omics and machine learning may enhance mechanistic and predictive insight, but only under rigorous quality control, validation, and reproducible pipelines. Overall, harmonized measurement, clear estimands, multi-ancestry replication, and integrated evidence pipelines are pivotal for producing robust and actionable G×D evidence.