Microarray Classification and Gene Selection with FS-NEAT

The analysis of microarrays has the potential to identify and predict diseases predisposition, such as cancer, opening a new path to better diagnosis and improved treatments. Additionally, microarrays can help to find genetic biomarkers, which are genes whose expressions are related to a specific disease stage or condition. But due to the huge number of genes present in microarray experiments, and the small number of available samples, computational methods that deal with such techniques need to overcome difficulties in both classification and feature selection tasks. This paper presents adaptations for the use of FS-NEAT, an evolutionary algorithm that creates and optimizes neural networks through genetic algorithms, as a tool that can satisfactorily perform both tasks simultaneously and automatically. The method is tested with a Leukemia dataset containing six imbalanced classes, compared with other classifiers, and the selected genes are biologically validated.