Evaluating advanced computing techniques for predicting breeding values in Harnali sheep.

Area of Science:

• Animal genetics and breeding
• Computational biology
• Statistical modeling

Background:

• Genetically superior animal conservation relies on accurate breeding value prediction.
• Advanced computing offers novel approaches to analyze complex animal population data.

Purpose of the Study:

• To evaluate artificial neural networks (ANN) and Bayesian techniques (BT) for predicting breeding values (BV) of weaning weight (WWT).
• To compare the prediction accuracy and model adequacy of ANN and BT in Harnali sheep.

Main Methods:

• Estimated BV for WWT using a univariate animal model and restricted maximum likelihood.
• Applied ANN (multilayer perceptron) and BT (Markov chain Monte Carlo) to predict BV on training and test datasets.
• Utilized goodness-of-fit criteria (R², RMSE, MAE, bias) and 10-fold cross-validation for evaluation.

Main Results:

• High prediction accuracy observed for both ANN (0.89) and BT (0.90) in predicting BV for WWT.
• Similar goodness-of-fit metrics and bias indicated comparable performance between ANN and BT.
• 10-fold cross-validation confirmed the analogous prediction abilities of both techniques.

Conclusions:

• Both ANN and BT demonstrate high capability and model adequacy for predicting breeding values.
• These advanced computational techniques can be effectively utilized in animal selection programs for genetic improvement.
• The study highlights the potential of AI and statistical modeling in enhancing livestock breeding strategies.