Fig. 4

True discovery versus false discovery curves comparing the performances of Multi-INTACT and INTACT for classifying gene-to-trait effects. Curves are truncated at a false discovery rate threshold of 0.20 for Multi-INTACT in order to represent the highest-ranked genes. Results represent 100 simulated data sets with causal genes from a variety of possible DAGs. For Multi-INTACT, we use gene product relevance probabilities \(P(E\ |\ \text {data})\) and \(P(P\ |\ \text {data})\) as scores for detecting \(E \rightarrow Y\) and \(P \rightarrow Y\) effects, respectively. We use the INTACT posterior, fastENLOC gene-level colocalization probability, and TWAS z-score magnitude with the respective molecular phenotype (expression, left; protein, right) as a classification score in each panel for comparison. For each curve, the false discovery rate threshold of 0.05 is denoted by an asterisk