Figure 3

The distribution of activity across descriptor spaces for compounds in the COX-2 data set. Symbols are color-coded by structural class and symbol sizes are proportional to the negative common logarithm of the potency (pIC50). Compounds falling into the three minor classes (cyclopentadienes, isoxazoles and thiophene DuP-697) are indicated in gray. Points in the vertical "hedges" at the top left and top right of the plots represent singletons that are too dissimilar to any other compound to be placed meaningfully within the eNLM. (A) Projection obtained by applying embedded non-linear mapping (eNLM) to the Euclidean distance matrix calculated from steric and electrostatic fields. Points reprsenting compounds from the minor classes are circled. (B) Projection obtained by applying eNLM to the Euclidean distance matrix calculated from molecular holograms hashed to a length of 353. See the Methods section for details.