bioresearch-ancestry-aware-mrlisted

# BioResearch Agent — Ancestry-Aware MR Skill ## Capability Runs cross-ancestry Mendelian randomization with pleiotropy-aware methods: 1. **Per-ancestry IVW** — inverse-variance-weighted MR per ancestry with Cochran's Q heterogeneity test, producing ancestry-specific causal effect estimates and confidence intervals. 2. **Cross-ancestry meta-analysis** — fixed-effects (FE) and random-effects (RE, DerSimonian-Laird) meta-analysis across ancestries, with Cochran's Q and I² statistics to quantify cross-ancestry heterogeneity. 3. **CAUSE-like model** — (Morrison et al. 2020) EM algorithm modeling both correlated and uncorrelated horizontal pleiotropy. Tests H0 (no causal effect, pleiotropy free to vary) vs H1 (causal effect + pleiotropy) via likelihood-ratio test. Distinguishes true causation from correlated pleiotropy — the key confound in standard MR. 4. **MRMix-like model** — (Wang et al. 2020) three-component mixture model (causal / pleiotropic / null) via EM, estimating the proportion of pleiotropic instruments and adjusting the causal estimate accordingly. 5. **Portability assessment** — evaluates whether the causal effect estimated in a reference ancestry (e.g., EUR) transfers to others: direction consistency, significance consistency, heterogeneity (I²), EUR-centric bias, and a composite transferability score (0–1). Returns per-ancestry MR + meta-analysis + CAUSE/MRMix + portability report, not a definitive causal claim. ## Run ```bash b