Accessing marker effects and heritability estimates from genome prediction by Bayesian regularized neural networks