Feature extraction from simulated datasets of molecular profiles using a Genetic Algoritm

Modern techniques of molecular biology may be used to classify an individual of unknown membership to one among the known groups of a reference population. DNA fingerprints, also called molecular profìles, provide an almost infinite amount of raw data for feature extraction, i.e. the Identification of molecular markers. Nevertheless, the huge search space and the small sample size of pilot experiments challenge common search algorithms. We numerically investigate the performances of a Genetic Algorithm (GA) that optimizes a fitness function based on predictive Bayesian distributions. Datasets of molecular profiles are simulated according to vectors of odds that fully characterize marginal distributions. Preliminary results from the analysis of simulations output give hints about the features of the reference population that increase the computational burden required for the Identification task.