Evaluating advanced computing techniques for predicting breeding values in Harnali sheep

Advanced computing techniques have been used by animal researchers to understand the intricate data structures for deriving the most reliable allusions of populations in order to conserve genetically superior animals. The present attempt was made to evaluate the potential of two advanced techniques, artificial neural networks (ANN) and Bayesian technique (BT), for predicting breeding values (BV) of weaning weight (WWT) using data of 498 lambs born to 41 sires and 173 dams in Harnali sheep for the period from 2014 to 2019. The estimated BV for WWT was initially obtained using univariate animal model under restricted maximum likelihood procedure. ANN using multilayer perceptron with two hidden layers was fitted to training set (75%) of estimated BV to predict BV for test set (25%). Similarly, BT using Markov chain Monte Carlo (MCMC) method was also fitted to similar datasets. The high accuracy of prediction, i.e., correlation between BV and predicted BV, was observed as 0.89 and 0.90 under ANN and BT, respectively. Further, similar ranges of goodness of fit criteria, viz., R-2, root mean square error (RMSE), mean absolute error (MAE), and bias, indicated that both ANN and BT had similar prediction ability, which was also confirmed by 10-fold cross-validation. The present study indicated high capability and analogous model adequacy for both techniques that can be exploited in selection programs.