Description
Mendelian Randomization Study
A two-sample Mendelian randomization study built from public genome-wide association summary statistics, executed to the standard that journals began enforcing once MR became easy enough for anyone to run. That shift is the whole point of this service. The software is free and the data are public, which produced a flood of single-exposure, single-outcome papers with no sensitivity analysis, and an editorial response that now rejects them on sight. What survives review is the version where every assumption is tested and reported.
New to the method? Start with our explainer on what Mendelian randomization is, when to use it, and how to conduct one, and the background on instrumental variable approaches to causal inference.
When This Is the Right Study
MR is the right instrument when you have a modifiable exposure with a well-powered GWAS, an outcome with an independent well-powered GWAS, and a plausible biological reason the exposure might cause the outcome. It is at its strongest for questions a randomised trial cannot ethically or practically answer, lifelong exposure to a biomarker, a nutrient, a body composition measure, and for triaging drug targets before expensive development.
It is the wrong instrument when the exposure has no strong genetic instruments, when the exposure and outcome GWAS come from overlapping samples, when the exposure is highly polygenic and biologically non-specific, or when the answer is already established. MR estimates a lifelong average effect, not the effect of an intervention at one point in life, and reviewers increasingly insist that papers say so.
How the Analysis Is Actually Done
The three instrumental variable assumptions, relevance, independence, and exclusion restriction, cannot all be proven. What a credible MR paper does is test each one as far as the data allow and report honestly where it cannot.
- Instrument selection stated and defensible – Genome-wide significant variants, clumped to independence at a stated linkage disequilibrium threshold and window, with the reference panel named. Where instruments are too few at conventional thresholds, a relaxed threshold is used only with the resulting weak-instrument bias quantified.
- Instrument strength quantified – F-statistics per instrument and overall, plus the proportion of exposure variance explained. Weak instruments in two-sample MR bias toward the null; in the presence of sample overlap they bias toward the confounded observational estimate. Both are reported, not glossed. We ran 18,000 simulated studies on this question: see what the F-statistic actually tells you.
- Harmonisation done carefully – Effect alleles aligned across studies, palindromic variants with intermediate allele frequency resolved or dropped, and the number dropped at each step reported. Silent strand errors are a leading cause of MR results that fail to replicate.
- Four estimators, not one – Inverse-variance weighted as the primary, with MR-Egger, weighted median and weighted mode as sensitivity analyses. These make different assumptions about pleiotropy; agreement across them is the evidence, and disagreement is reported rather than buried. Our simulation study sets out what each of the four estimators actually protects you from, and where each one breaks.
- Directional pleiotropy tested – MR-Egger intercept with its interpretation, plus the I-squared-GX statistic for the NOME assumption, since a low value makes the Egger estimate itself unreliable.
- Outliers detected formally – MR-PRESSO global test, outlier test and distortion test, with results reported before and after outlier removal.
- Heterogeneity quantified – Cochran’s Q with leave-one-out analysis, so no single variant is driving the finding.
- Reverse causation excluded – Steiger filtering to confirm each instrument explains more variance in the exposure than in the outcome, plus bidirectional MR where the direction is genuinely in question.
- Instrument PheWAS – Each instrument checked against phenome-wide association resources for alternative pathways to the outcome. This is where exclusion-restriction violations actually get caught, and it is the step most submitted MR papers skip.
- Independent replication – The analysis repeated in a second outcome GWAS from a different cohort wherever one exists.
- Ancestry matched – Exposure and outcome GWAS drawn from the same ancestral population, with population stratification addressed explicitly.
What You Receive
- Complete manuscript – Written to STROBE-MR reporting standards for your target journal
- Full figure set – Scatter plot with all estimator slopes, forest plot of per-instrument effects, funnel plot, and leave-one-out plot, as publication-quality vector output
- Instrument table – Every variant with effect size, standard error, effect allele frequency, F-statistic and PheWAS annotation
- Full analysis code – R using TwoSampleMR and MRPRESSO, commented and runnable end to end from public summary statistics
- Completed STROBE-MR checklist
- One round of author-requested revisions
What Is Not Included
We do not ghost-write authorship. You set the hypothesis, review the biological interpretation, and own the claims. We do not perform individual-level genetic analysis, GWAS of your own cohort, or anything requiring access to a restricted biobank under your credentials. We do not submit for you and do not guarantee acceptance. If the causal estimate is null, the manuscript reports a null, a well-executed null MR is publishable and we will write it that way.
When We Will Tell You Not to Buy This
If your exposure has fewer than roughly ten independent genome-wide significant instruments, the analysis will be underpowered and the sensitivity estimators will be uninformative. If the only available exposure and outcome GWAS share a substantial sample, the result will be biased and no method fixes it. If the same exposure-outcome pair has already been published from the same summary statistics, a duplicate adds nothing. We check the instrument count, the F-statistics, the sample overlap and the existing literature before starting, that check is what the Feasibility & Protocol Assessment is for.
Data Access
Public GWAS summary statistics. We obtain them ourselves from the standard repositories, there is no application and nothing for you to arrange. Where a consortium requires a brief data request, we will tell you at the outset whether it needs your signature.
Timeline
Standard delivery is 15 business days from an agreed analysis plan. Multivariable, bidirectional and colocalisation extensions take longer; if yours will, we say so before you pay.
Common Questions
What if the estimators disagree? That is a finding, not a failure. Disagreement between IVW and the pleiotropy-robust methods tells you something about the instruments, and the manuscript reports and interprets it.
Can you do drug-target MR? Yes. Instruments restricted to the gene region encoding the target, with colocalisation to confirm a shared causal variant. This is quoted separately.
Is MR strong evidence? It sits above conventional observational epidemiology and below a randomised trial. See our guide to levels of evidence in clinical research.
My MR paper was desk-rejected. Can you rescue it? Often. Most desk rejections in this area are for missing sensitivity analyses rather than a wrong core result. Send the manuscript.
Extended Scope
Multivariable MR for correlated exposures, mediation MR, colocalisation, drug-target MR and non-linear MR are quoted individually, typically from $3,500. See the Extended & Multi-Database Registry Study.
Have a Question First?
If you are not sure this is the right fit for your research question, ask before you buy. Send it below and a senior analyst replies within one business day. There is no obligation, and if a different product suits you better, or if the honest answer is that we cannot help, we will say so.
