New technologies such as combinatorial chemistry and automated high-
hroughput screening of multiple biological targets have revolutionized
the drug discovery process.  Additionally the desire to utilize large
collections of patient data combined with genetics information will
soon drastically change how clinical trials will be performed and
assessed.  Techniques for analyzing these data are not well defined.
For new biological targets it is important to determine which compounds
should be screened first and determine the features of small molecules
important for biological activity.  For clinical trials it is important
to determine both clinical and genetic factors that affect both efficacy
and safety.  Recursive partitioning is a very effective method for
summarizing and finding trends in very large data sets. We apply this
data mining technique to identify active chemical subclasses during
high-throughput screening, defining subpopulations in large patient
databases, and untangling complex gene interactions.