Can neural network constraints in GP provide power to detect genes associated with human disease?

A major goal of human genetics is the identification of susceptibility genes associated with common, complex diseases. Identifying gene-gene and gene-environment interactions which comprise the genetic architecture for a majority of common diseases is a difficult challenge. To this end, novel computational approaches have been applied to studies of human disease. Previously, a GP neural network (GPNN) approach was employed. Although the GPNN method has been quite successful, a clear comparison of GPNN and GP alone to detect genetic effects has not been made. In this paper, we demonstrate that using NN evolved by GP can be more powerful than GP alone. This is most likely due to the confined search space of the GPNN approach, in comparison to a free form GP. This study demonstrates the utility of using GP to evolve NN in studies of the genetics of common, complex human disease.