Integrative Mendelian randomization for detecting exposure-by-group interactions using group-specific and combined summary statistics

This Zenodo archive provides the dataset underlying the manuscript currently under revision at PLOS Genetics. Full code and documentation, along with the accompanying software implementations and scripts used to reproduce the analyses and generate all figures in the manuscript, are maintained in the project’s GitHub repository https://github.com/kxu-stat/int2MR. Abstract Interactions between risk factors and covariate‑defined groups are commonly observed in complex diseases, yet existing methods for detecting such interactions typically require individual‑level data. Data availability and measurements of risk exposures and covariates often limit statistical power and applicability. To overcome these challenges, we propose int2MR, an integrative Mendelian randomization (MR) method that leverages GWAS summary statistics—both group‑separated and group‑combined—for outcome traits in conjunction with exposure GWAS results. By integrating additional group‑combined data, int2MR can assess a broad range of exposure effects and reveal interaction signals that are unattainable using incomplete individual‑level datasets alone. Simulation studies demonstrate that int2MR effectively controls type I error across diverse scenarios while achieving substantial power gains. In real‑world applications, we first identify risk exposures with sex‑interaction effects on ADHD—our findings suggest elevated inflammatory processes in males—and then detect age‑group‑specific risk factors for Alzheimer’s disease pathologies in the oldest‑old (age 95+), many of which implicate immune and inflammatory pathways. Overall, int2MR offers a robust and flexible framework for uncovering group‑specific and interaction effects, yielding new insights into disease mechanisms. Contents of this Archive. The file int2MR_ADHD_results provides both group-specific and interaction causal effect estimates generated by our int2MR method, directly supporting the section “Data Analysis: Identifying Risk Factors with Sex-Biased Effects on ADHD”. These results include all parameter estimates and p-values necessary to reproduce the figures and tables discussed in that part of the manuscript. The file int2MR_AD_results provides both group-specific and interaction causal effect estimates generated by our int2MR method, directly supporting the section “Data Analysis: Identifying Age‑Group‑Specific Risk Factors for Alzheimer’s Disease in the Oldest‑Old.” These results include all parameter estimates, standard errors, and p-values necessary to reproduce the figures and tables discussed in that part of the manuscript. The compressed folder AD_ROSMAP contains instrumental‑variable–to–outcome effect estimates derived from the ROSMAP study. This ZIP file comprises two subfolders, rosmap_95p (participants aged > 95) and rosmap_all (the entire cohort), each holding GWAS summary statistics. For example, the file tangles.assoc.linear.gz reports PHF tau tangle density associations from linear‐regression analyses (binary traits use logistic regression). Readers may consult the “Data Analysis: Identifying Risk Factors with Sex‑Biased Effects on ADHD” section of the main text for further methodological details. If you have any questions or need assistance reproducing these results, please contact Ke Xu at [e-mail].