Figure 3.

Performance testing strategy of GESS methods. (A) The GESs of the drugs in each MOA and SSC category were searched against the LINCS database with each of the six GESS methods. The results were sorted by the corresponding similarity scores, here indicated by boxes with color gradient. GESs from the same and different MOA/SSC categories (CAT) as the query were indicated in a binary vector with ones and zeros (next to boxes), respectively. After joining the binary vectors for each category group and re-sorting them by the corresponding scores, cumulative TPRs and FPRs were plotted in form of ROCs. This was done on the global level (B) and the CAT level (C) for the MOA and SSC classifications separately. (D) The distributions of AUC/pAUC values from each CAT-level are depicted by violin plots with mean values and standard deviation (STDEV) bars given in the middle. In addition, the global AUC/pAUC values are indicated by triangles. (E) The statistical significance of the observed differences among the global AUC/pAUC values of the six GESS methods was assessed by a bootstrap test described in the text.