MR-Clust

MR-Clust identifies clusters of genetic variants with similar causal estimates to characterize clustered heterogeneity in Mendelian randomization and infer distinct causal pathways between risk factors and outcomes.

Key Features:

• Clustered Heterogeneity Detection: Identifies clusters of genetic variants with similar ratio-estimates in direction, magnitude, and precision, reflecting distinct causal pathways.
• Expectation-Maximization (EM) Algorithm: Fits a mixture model using an EM-based model fitting approach to assign variants to clusters while accounting for uncertainty in causal estimates.
• Inclusion of Null and Junk Clusters: Incorporates null and "junk" clusters to reduce detection of spurious clusters and improve robustness.
• Superior Performance in Simulations: Demonstrates superior detection of the number of clusters in simulation studies compared with methods such as Mclust.
• Application Example: Applied to the effect of blood pressure on coronary artery disease, identifying four clusters including one with a negative causal effect associated with trunk fat percentage and other adiposity measures.

Scientific Applications:

• Epidemiological Causal Inference: Dissects causal mechanisms in Mendelian randomization studies by grouping variants into mechanistic clusters.
• Pathway and Target Discovery: Facilitates identification of distinct genetic pathways and potential therapeutic targets by separating variant clusters with different causal effects.
• Handling Heterogeneity and Pleiotropy: Enables analysis of clustered heterogeneity to resolve complex relationships between genetic variants and disease phenotypes.

Methodology:

MR-Clust assesses causal estimates from individual genetic variants, groups variants by similarity in causal (ratio) estimates using a mixture model, fits the model with an EM algorithm, and includes null and junk clusters to guard against spurious findings.