For each compound, we computed the true positive rate (i.e., recovery of expected relationships). We refer to these expected relationships as expected pairs. To estimate the false positive rate, we counted the connections between compounds and genetic or pharmacologic perturbagens annotated as having a relationship with a different small molecule in the dataset. We refer to such relationships as null pairs. We then plotted the true positive rate against the false positive rate at various thresholds of statistical significance, thereby generating an ROC curve from which an AUC could be calculated. An AUC >0.6 is typically regarded as signifying a positive signal. At that cut-off, an average of 45% of expected relationships were recovered in any one of the 9 cell lines tested (range 29%-58%). This number rose to 63% when Connectivity Scores were summarized across all 9 lines (see STAR Methods, Table S6).
