A Bayesian Algorithm for In Vitro Molecular Evolution of Pattern Classifiers

We use molecular computation to solve pattern classification problems. DNA molecules encode data items and the DNA library represents the empirical probability distribution of data. Molecular bio-lab operations are used to compute conditional probabilities that decide the class label. This probabilistic computational model distinguishes itself from the conventional DNA computing models in that the entire molecular population constitutes the solution to the problem as an ensemble. One important issue in this approach is how to automatically learn the probability distribution of the DNA-based classifier from observed data. Here we develop a molecular evolutionary algorithm inspired by directed evolution, and derive its molecular learning rule from Bayesian decision theory. We investigate through simulation the convergence behaviors of the molecular Bayesian evolutionary algorithm on a concrete problem from statistical pattern classification.