Evolutionary Optimization of A Neural Network-Based Signal Processor for Photometric Data from An Automated DNA Sequencer

We are using signal conditioning neural networks to improve DNA sequencing. In the course of exploring how to improve the efficiency of these neural networks we have looked at evolutionary computation as a method of optimization. This paper explores genetic algorithms as a means of discovering large arrays of real numbers such as those found in a neural network weight matrix. A new genetic component, the intron, is introduced as a way of preserving genetic variability in the presence of higher crossover and mutation rates. However, the introduction of introns into a representational architecture increases the length of the chromosome. A method for systematically exploring the changing chromosomal length and how that influences the genetic algorithms control parameters is discussed in terms of a simulated problem. Finally, a photometric perceptron-signal filter is evolved using the best parameters discovered during the course of this research.