Abstract:Background and Aims: Dietary factors may contribute to the development of pancreatic cancer, but conventional observational studies are susceptible to confounding, measurement error, and reverse causation. Food liking is an upstream behavioral phenotype that influences long-term dietary behavior and has a genetic component, yet its association with pancreatic cancer risk has not been systematically evaluated. This study investigated the potential associations between genetically predicted food liking phenotypes and pancreatic cancer risk using a genetic epidemiological approach. Methods: A two-sample Mendelian randomization (MR) design was employed. Exposure data were obtained from a genome-wide association study (GWAS) of food liking conducted by May-Wilson et al. in European-ancestry participants from the UK Biobank, covering 187 food liking phenotypes. Outcome data were derived from three European-ancestry pancreatic cancer GWAS datasets, with FinnGen serving as the primary dataset and ukb-saige-157 and ebi-a-GCST90018893 as supplementary replication datasets. The inverse-variance weighted (IVW) method was used for the primary MR analysis, complemented by MR-Egger regression, weighted median, weighted mode, and multiple sensitivity analyses. Candidate associations identified in the primary analysis were further evaluated by external replication, random-effects meta-analysis, and correction for potential sample overlap. Results: In the primary FinnGen analysis, 11 food liking phenotypes showed nominally significant associations with pancreatic cancer risk. After external replication and meta-analysis, eight phenotypes remained statistically associated with pancreatic cancer risk. Genetically predicted liking for chilli pepper and grapefruit was associated with lower pancreatic cancer risk, with pooled OR (95% CI) of 0.83 (0.74-0.93) and 0.89 (0.81-0.99), respectively, whereas milky sweets liking was associated with higher risk, with a pooled OR of 1.15 (1.04-1.26). The other five meta-analysis-supported associations were considered exploratory because of limited support in individual external datasets. Three phenotypes that showed nominal associations in the primary analysis were not supported by meta-analysis. Sensitivity analyses revealed no substantial evidence of heterogeneity, directional horizontal pleiotropy, or single-SNP-driven effects. MRlap analysis indicated that correction for potential sample overlap did not materially alter the effect estimates. Conclusion: Genetic epidemiological analyses suggest that several food liking phenotypes may be genetically associated with pancreatic cancer risk. Liking for chilli pepper, grapefruit, and milky sweets showed relatively consistent association signals. Because food liking does not directly represent actual food intake, these findings do not provide sufficient evidence to support dietary recommendations targeting specific foods. Further validation is warranted in larger samples, populations of diverse ancestries, and studies incorporating more refined dietary behavioral phenotypes.