If desired, certain properties of the selected compounds can be optimized along with their diversity. This is done by defining a set of property filters (0 ≤ LogP ≤ 5, 200 ≤ MW ≤ 500, etc.) and biasing the selection of compounds to satisfy as many filters as possible. Failing a given filter does not disqualify a compound, but a compound that fails too many filters will be detrimental to the property score and will tend not to be selected. This approach allows the overall properties of the selected compounds to be improved, without summarily disqualifying a large fraction of the pool from which selections are made.
A variety of Canvas tools can be employed to verify that the compounds being selected are highly diverse and are in fact filling holes in the reference library. Figure 2a contains a distribution of nearest neighbor Tanimoto similarities within a reference library of 1000 compounds from the Asinex Platinum Collection.2 Figure 2b shows how the distribution shifts dramatically to lower similarities after using Canvas Hole-Filling to add 1000 more Asinex compounds to the reference library from a pool of 5000. Figure 2c illustrates that adding 1000 random compounds to the reference library from the same pool makes very little impact on the shape of the distribution. This confirms that the shift in Figure 2b is not simply a consequence of there being higher diversity within the pool from which selections were made.
Figure 2: Distributions of nearest neighbor Tanimoto similarities for compounds from the Asinex Platinum Collection are shown above for (a) a reference library of 1000 compounds; (b) the reference library plus 1000 compounds chosen to fill holes; and (c) the reference library plus 1000 randomly chosen compounds.