Fig. 3

Correlation between preferences and the mapping of current checkmol fingerprints. (A) The mapped region of checkmol fingerprint and their odds-ratio ranking (epsilon = 0.5, is added to prevent dividing by zero) of 204 checkmol descriptors within each family of the PKFIs set. Most pre-moieties associated with different modes of kinase inhibition (described previously) are mappable and correlated with the feature distribution in the training sets. However, several pre-moieties are still not precisely defined in the current fingerprint and thus are undistinguishable (i.e., naphthalenyl-nitrogen regions on lapatinib). (B) Overall odds-ratio distribution of checkmol descriptors on EGFR, JAK, and PIM datasets with mapped moieties is indicated. It should be noted that several super peaks are observed with relatively few compounds and thus are not currently discussed