Fig. 3: High-dimensional mediation.
From: Transcripts with high distal heritability mediate genetic effects on complex metabolic traits
A Workflow indicating major steps of high-dimensional mediation. The genotype, transcriptome, and phenotype matrices were kernelized to yield single matrices representing the relationships between all individuals for each data modality (GK = genome kernel, TK = transcriptome kernel; PK = phenome kernel). High-dimensional mediation was applied to these matrices to maximize the direct path G → T → P, the mediating pathway (arrows), while simultaneously minimizing the direct G → P pathway (dotted line). The composite vectors that resulted from high-dimensional mediation were GC, TC, and PC. The partial correlations ρ between these vectors indicated perfect mediation. Transcript and trait loadings were calculated as described in the methods. B The null distribution of the path coefficient derived from 10,000 permutations. Comparisons are shown to the observed path coefficient (red) the path coefficient using a distal-only model (gray) and the path coefficient using the local-only model (brown). C. The null distribution of the GC-TC correlation vs. the TC-PC correlation. Comparisons are shown to the observed values (red), and those derived from the distal-only model (gray) and the local-only model (brown). Source data are provided as a Source Data file.
