For example, we used the annotation of protein targets of small molecule drugs and tool compounds to determine whether such targets could be recovered from the CMap. Our analysis showed that the CMap results were highly enriched in the correct targets for up to 63% of small molecules tested. While this result is encouraging, 37% of compounds showed no evidence of connection to their expected targets. Failure to recover such connections could be explained by many factors including i) incomplete inhibition of the target by the compound, ii) off-target effects of compounds and genetic perturbations, iii) missing information in the L1000 read-out, iv) incorrect literature-based annotations of compounds, v) biological differences between small molecule inhibition of specific aspects of protein function (e.g., enzymatic inhibition) compared to complete loss of function (e.g., scaffolding functions) induced by shRNA-mediated knock-down, and vi) the existence of previously unrecognized bona fide connections that effectively penalize the known connections – particularly if the novel connections are stronger than the expected ones.
