Improving genetic programming for the prediction of pharmacokinetic parameters

The prediction of pharmacokinetic parameters is a crucial phase of the drug discovery process, and the automatization of this task is a hot topic in computational biomedicine. In the last 10 years, a significant amount of research has been published reporting on applications of genetic programming to the prediction of pharmacokinetic parameters. This paper summarizes and discusses some of those contributions. In particular, the focus is on the idea that pharmacokinetic problems are so complex that the "canonic" version of genetic programming is often not able to perform appropriately on them. At the same time, genetic programming has a high degree of versatility, given by the opportunity it offers of adapting many crucial parts of its algorithm, among which the fitness function and the employed genetic operators. This gives us the chance to improve standard genetic programming in several different ways. For instance, sophisticated fitness functions, methods to control bloat and operators to exploit the geometry of the semantic space are discussed here.