Genetic programming models for classification of data from biological systems

Data classification problems especially for biological systems pose serious challenges mainly due to the presence of multivariate and highly nonlinear interactions between variables. Specimens that need to be distinguished from one another show similar profiles and cannot be separated easily based on decision boundaries or available discriminating rules. Alternatively, inter-relations among the feature vectors can be exploited for distinguishing samples into specific classes. Such variable interaction models are difficult to establish given the nature of biological systems. Genetic Programming, a data driven evolutionary modeling approach is proposed here to be a potential tool for designing variable dependency models and exploiting them further for class discrimination. A new and alternative GP model based classification approach is proposed. Analysis is carried out using illustrative datasets and the performance is benchmarked against well established linear and nonlinear classifiers like LDA, kNN, CART, ANN and SVM. It is demonstrated that GP based models can be effective tools for separating unknown biological systems into different classes. The new classification method has the potential to be effectively and successfully extended to many systems biology applications of recent interest.