Database transformations are a frequent problem for data managers supporting scientific databases, particularly those connected with the Human Genome Project. The databases involved frequently contain complex data structures not typically found in conventional databases, such as arbitrarily nested records, sets, variants, and optional fields, as well as object identities and recursive data structures. Furthermore, programs implementing the transfomations must be frequently modified since the databases involved evolve rapidly, as often as 3 to 4 times a year. We present in this paper a language (WOL) for specifying transformations between such databases and describe its implementation in a system called Morphase. Optimization are performed at all stages, with significant impact on the compilation and execution time of sample transformations.