Hello.

Genome-wide association study summary statistics, which include effect sizes, standard errors, allele information, and case and control numbers, are sufficient to conduct a robust two-sample Mendelian randomization analysis between two diseases. Individual-level genotype data is not required for two-sample Mendelian randomization, as this method relies on publicly available summary data from separate genome-wide association studies for the exposure and outcome.

The key steps and best practices to ensure the validity of your analysis are as follows:

• Select genetic variants as instruments that are strongly associated with the exposure disease, typically using a genome-wide significance threshold of p-value less than 5 times 10 to the power of minus 8, and ensure instrument strength with an F-statistic greater than 10 to avoid weak instrument bias.
• Perform linkage disequilibrium clumping to select independent instruments, using a clumping window of 10,000 kilobases and an r-squared threshold of 0.001, based on a reference panel such as the 1000 Genomes European ancestry data.
• Harmonize the exposure and outcome data by aligning alleles, ensuring the effect allele is consistent across datasets, and removing ambiguous palindromic variants if allele frequencies are not available to infer strand.
• Apply Mendelian randomization methods such as inverse-variance weighted, weighted median, and Mendelian randomization-Egger to estimate causal effects, with inverse-variance weighted as the primary method.
• Test for horizontal pleiotropy using Mendelian randomization-Egger intercept test and MR-PRESSO, and assess heterogeneity with Cochran's Q statistic.
• For binary traits, transform effect sizes to the liability scale if needed for interpretation, using formulas that account for disease prevalence, though this is optional depending on whether you aim to report odds ratios or absolute risks.
• Conduct sensitivity analyses, including leave-one-out analysis to identify influential variants, and use methods like Mendelian randomization mode-based estimate for robustness.
• Ensure no sample overlap between the exposure and outcome genome-wide association studies to avoid bias from weak instruments or winner's curse.

Recommended resources include the TwoSampleMR package in R, which is actively maintained and supports all major Mendelian randomization methods; its vignette provides a comprehensive tutorial at https://mrcieu.github.io/TwoSampleMR/. A useful tutorial on Mendelian randomization with binary traits is available from the Medical Research Council Integrative Epidemiology Unit at the University of Bristol. Example studies include the 2020 paper by Burgess et al. in the International Journal of Epidemiology on Mendelian randomization methods for binary outcomes, and a 2022 study by Yuan et al. in Nature Genetics applying two-sample Mendelian randomization to investigate causal relationships between autoimmune diseases.

Kevin