Peptide Classification with Genetic Programming Ensemble of Generalised Indicator Models

The generalised indicator model (GIM) has been developed for peptide classi square cation with success. However, the performance of GIM varies with the mutation matrix which is used to measure the similarity between peptides. This work investigates three methods for building meta-classi square ers based on GIMs which are treated as base classi square ers constructed using different mutation matrices. The three methods are linear combination, neural network combination and genetic programming. The simulation shows that the genetic programming method performs the best in two aspects. First, it is able to identify the most important base classi square ers for building a meta-classi square er without any a priori knowledge. Second, a metaclass square er delivered is a mathematical equation being capable of interpretation.