Fig 1.
(A) Analysis workflow. Blue rhomboids are data inputs and green rhomboids are analysis outputs. MAF, Minor Allele Frequency. HWE, Hardy-Weinberg Equilibrium. (B) Hierarchy of functions increasing in specificity from top to bottom. Color shows strength of association. (C) Dimensionality reduction and association workflow. For each system defining a set of genes (left), a SNP matrix (center) is constructed from variants near system genes (blue bands in underlying chromosomes). Matrix columns: SNPs; Rows: Individuals. Values (0, 1, 2) are (major allele homozygous, heterozygous, minor allele homozygous) at each SNP. A low-dimensional representation is computed using SVD (center right) and tested for association with phenotype (model f1) in comparison to the null hypothesis (f0).
Fig 2.
Subhierarchy of significantly associated systems.
System color corresponds to unadjusted p-value. Systems with FWER < 0.05 are highlighted with red outline.
Fig 3.
Association results for the Vesicle Transport Along Microtubule system (GO:0047496).
(A) Manhattan plot of the standard GWAS results for the primary COPD cohort. SNPs assigned to genes associated with GO:0047496 are highlighted in red. Horizontal line marks the standard GWAS significance threshold, p = 5 × 10−8. (B) Contributed effects of SNPs in the principal component regression model (system genetic components) for GO:0047496. SNPs are sorted by genomic position. Horizontal distance between SNPs does not correspond to genomic distance. SNP effect is the change in log odds ratio per minor allele copy contributed to the principal component regression for the entire system.
Table 1.
Top ten systems most significantly associated with COPD status.
Fig 4.
Conditional association test results.
(A) Comparison of association results for the original uncorrected analysis (red) and conditional tests (blue, green) for the top 10 most associated systems. (B) Conditional test results for the vesicle transport subhierarchy. P-values are unadjusted. Shaded bars for the “top child” test indicate the system had no children and thus no correction was applied.
Table 2.
Number of significant systems by each method for different significance criteria.