(A) An overview of GEMOT for the scenario depicted in Figure 1—figure supplement 1A. GEMOT incorporates 3 stages (I, II and III) that are schematically described in B–D, E–F and G–I, respectively. Stage I: (B) High and low link-potential scores for pairs of variants and traits. A typical calculation for variant–trait pairs (left: v1–p1, right: v1–p2). Shown are variant–transcript associations (y-axis) and transcript–trait correlations (x-axis) for each transcript (black dots). GEMOT uses a threshold (horizontal dashed line) to compare the distribution of transcript–trait correlation in high and low transcript–variant associations scores. Link potential is high (left) when the distributions of correlation differ between the high and low association range more than expected by chance; link potential is low (right) where no difference is observed. Notably, a high link potential reflects the potential that some transcripts may translate between a variant and a trait, although such transcripts are not yet specified. (C) Bipartite graph construction. GEMOT constructs a graph whose two parts are variants (squares) and traits (diamonds); edge weights are the link-potential scores (solid lines, high; dashed lines, low). (D) Bipartite module identification. GEMOT identifies ‘heavy’ biclusters in the bipartite graph (in this example, 1 module). Stage II: (E) Transcript link score. The input is provided by all calculated correlation and association scores (such as the three plots on the left). On the right: given a transcript, GEMOT aggregates and ranks all its scores in a horizontal track (red, correlations; blue, associations) and uses the distribution of ranks to score the transcript for significance. (F) Tripartite module construction. GEMOT adds high-scoring transcripts from E, referred to as ‘candidate transcripts’ (circles), to the module. Stage III: (G) Causality p value scores. GEMOT uses a statistical score to assess causative relationships (blue, significant; white, non-significant) for each transcript (row) and trait (column) in a module. Non-causative relationships attain non-significant scores (cartoon examples on the left). (H) Module refinement. Starting with the causality p value scores for the tripartite module (from G), GEMOT eliminates non-causative transcripts (left) and non-affected traits (right) in an iterative manner. (I) The resulting GEMOT module (arrows, causative relationships).