Influence of Gene Parameter Nomorlization on Back-Propagation Artificial Neural Network Caculating Time of Animal Phenotype Value Prediction

The complex Interconnections between markers and polygenic genotype value suggested that the regression was not enough for describing the relation between genes and traits. Artificial neural networks (ANNs) could perform well for optimization in complex non-linear systems. Recently, artificial neural networks had been successfully used to predict the polygenic genotype value, and the different learning rate and hidden neurons number were used to discuss the influencing of the learning rate on estimating the polygenic genotype value. However, when optimazing the structure of BP-artificial neural networks, a series of networks with an variable number of hidden neurons and input neuron needs to be optimized, compared and selected, the elapsed time could be very long, therefore the elapsed time was very important for work efficients. In this paper, the influence of different gene parameter nomorlization on back-propagation artificial neural network caculating time of animal phenotype value pretiction was discussed. The results showed that the caculating time could be affected by many gene parameters, such as the gene effect, gene locus number, gene frequency, and their nomorlization, the normorlization can improve the training speed and induce absolute time or elapsed time very obviously. These suggested that normalizing the phenotype value was an very important method for improving our work efficient.