Having demonstrated the ability to recover MOA from optimized drugs and tool compounds, we next asked whether CMap could identify the MOA of previously uncharacterized compounds. Projection of TAS-high compounds in two dimensions shows that many uncharacterized compounds cluster with existing PCLs (Figure 5C). We focused on novel kinase inhibitors simply because of the availability of methods for validating CMap predictions. For example, our analysis indicated that the unannotated compound BRD-2751 showed strong connectivity to the Rho-associated protein kinase (ROCK) PCL, suggesting that it might in fact be a ROCK inhibitor. To test this hypothesis, we subjected the compound to kinome-wide binding measurements (using the Kinomescan assay) and found that precisely as predicted, the compound has a KD of 56 nM against ROCK1 (Figure 6A). We note that while the compound had not been previously reported to be a ROCK1 inhibitor, its chemical structure is reminiscent of canonical ROCK inhibitory compounds. As another example, several compounds (BRD-5161, BRD-5657, and BRD-9186) were predicted to function as MTOR and/or PI3 kinase inhibitors. Kinomescan dose-response profiling confirmed that the three compounds were indeed MTOR/PI3K inhibitors, spanning a range of potencies and selectivities (Figure S3B).
