Fig. 4: Colocalization analysis and pathway analysis.

The heatmap shows the overall Bayesian colocalization (‘coloc’ R package) results between different first-order features (a) or shape features (b) and significant eQTL genes, which applies a Bayesian framework to estimate posterior probabilities for different hypotheses. For each radiomic feature-eQTL gene colocalization result, the colocalization posteriori probability of hypothesis 4 (PPH4) is indicated with shades of red (closer to 1) and blue (closer to 0). Results with PPH4 > 0.7 are marked with ‘*’, and results with PPH4 > 0.9 are marked with ‘**’. For the colocalization results of a phenotype across the five lung lobes, the result with the highest PPH4 is shown. The venn diagram illustrates the overlap of candidate genes identified by the four methods (Near gene, SuSIE, coloc, MAGMA) for first-order features (c) and shape features (d). Pathway enrichment analyses of Gene Ontology terms were performed using ‘clusterProfiler’ R package. For each GO term, a hypergeometric test (two-sided) was used to assess over-representation of candidate genes (identified from SuSIE, colocalizations analysis, positional mapping, and MAGMA) for first-order features (e) and shape features (f). P values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR < 0.05). The enrichment of the top 15 pathways was visualized. posmap position mapping, SuSIE sum of single effects, coloc colocalization, MAGMA multi-marker analysis of genomic annotation.