A Genetic Algorithm Approach for Discovering Diagnostic Patterns in Molecular Measurement Data

The objective of this work is the development of an algorithm that, after training, will be able to discriminate between disease classes in molecular data. The system proposed uses a genetic algorithm (GA) to achieve this discrimination. We apply our method to three publicly available data sets. Two of the data sets are based on microarray data that allow the simultaneous measurement of the expression levels of genes under different disease states. The third data set is based on serum proteomic pattern diagnostics of ovarian cancer using high-resolution mass spectrometry to extract a set of biomarker classifiers. We show how our methodology finds an abundance of different feature models, automatically selecting a subset of discriminatory features, whose classification accuracy is comparable to other approaches considered. This raises questions about how to choose among the many competing models, while simultaneously estimating the prediction accuracy of the chosen models.