MUG '00 -- 14th Daylight User Group Meeting -- 22-25 Feb 2000
Dupont Pharmaceutical Research Labs (ne' CombiChem)
Daylight Chemical Information Systems, Inc.
A novel method for designing combinatorial libraries has been developed. It is based on the hypothesis that similar molecules have a higher probability to bind to the same receptor than average. For a given target, a training set of known actives and inactives is the starting point. Virtual libraries are defined by a reaction transformation and lists of reagents. Virtual molecules are first enumerated and then scored. The score is the average activity of the "N" nearest neighbours (in Tanimota space) in the training set.
The new method encoded in the program "like" works as follows: The training set of molecules with known activities is stored in a merlin database. "like" reads one virtual molecule at the time from a tdt file and adds the score by averaging the activity of the "N" nearest neighbours in the merlin database. The output tdt file can then be loaded in merlin, sorted by score and additional selection criteria can be applied. The result is a list of molecules suggested for synthesis. Such a method combines well with the high throughput synthesis system we built in San Diego. It has been optimized to perform well for "cherry-picked" compounds (as opposed to array synthesis), which is the typical characteristic of individually designed compounds. Some early applications of this method to ongoing drug discovery projects will be discussed.
Daylight Chemical Information Systems, Inc. , firstname.lastname@example.org