bio-causal-genomics-pleiotropy-detection

What it does

Detect and adjust for horizontal pleiotropy in two-sample Mendelian randomization by distinguishing uncorrelated (UHP) from correlated (CHP) pleiotropy and selecting among multiple pleiotropy-robust methods. It supports evaluating pleiotropy, performing a method battery, and handling scenarios with weak instruments or polygenic exposures.

How it works

The skill instructs the agent to:

• Use two-sample MR with methods: IVW, MR-Egger, weighted median, weighted mode, MR-PRESSO, CAUSE, LHC-MR, LCV, MR-Clust, MR-Mix, and contamination mixture, depending on the regime.
• Decompose pleiotropy into UHP (addressed by Egger / median / mode / MR-PRESSO) and CHP (addressed by CAUSE / LHC-MR / LCV).
• Run a standard battery: compute point estimates, SE, p, 95% CI, and n_SNPs_used for IVW, MR-Egger, weighted median, weighted mode, and MR-PRESSO (with NbDistribution >= 10000 for reporting).
• Use CAUSE for CHP-aware estimation or MR-Clust for mechanism-heterogeneous instruments; use MR-ConMix or MRMix for contamination or mixture approaches.
• Compute Egger I^2_GX and apply SIMEX if I^2_GX < 0.9; otherwise consider alternative methods (MR-RAPS).
• Follow an operational decision flow: compute genetic correlation with LDSC; escalate CHP if rg is high or biology suggests shared factors; apply a triangulated approach with Steiger checks and bidirectional MR.
• Report ELPD delta and CHP-specific metrics when escalating to CHP methods.

When to use it

Use when validating an MR causal claim, running the STROBE-MR sensitivity battery, suspecting a shared heritable confounder, working under weak-instrument or polygenic-exposure regimes, or reconciling discordant estimates across robust methods.

What it can touch

The skill references tools and packages in R: TwoSampleMR, MR-PRESSO, cause, mrclust, MendelianRandomization, MRMix, simex. It lists specific function calls and package names as examples for various methods:

• TwoSampleMR::mr(), mr_pleiotropy_test(), mr_heterogeneity(), mr_leaveoneout(), directionality_test
• MR-PRESSO::mr_presso()
• cause::cause()
• mrclust::mr_clust_em()
• MendelianRandomization::mr_conmix()
• MRMix::MRMix()
• MR-Mix, contamination-mixture, LHC-MR, LCV

Caveats

Imposes version compatibility notes (TwoSampleMR, MendelianRandomization, MR-PRESSO, CAUSE, MR-Clust, MRMix, simex) and requires careful interpretation when CHP is plausible or when fewer than 100 significant SNPs are available for CAUSE. For CHP-focused evaluation, CAUSE or LHC-MR are recommended; PRESSO may be underpowered under CHP. For few SNPs, avoid relying solely on MR-Egger; consider alternative methods and report the intercept with caveats.