Dash Spandan Shashwat, Bangar Yogesh C, Magotra Ankit, Patil C S
Department of Animal Genetics and Breeding, Lala Lajpat Rai University of Veterinary and Animal Sciences, Hisar, Haryana, 125004, India.
Directorate of Research, Sher-e-Kashmir University of Agricultural Sciences and Technology of Jammu, (SKUAST Jammu), Jammu, Kashmir, 180009, India.
Trop Anim Health Prod. 2025 Jun 3;57(5):241. doi: 10.1007/s11250-025-04495-4.
Machine learning techniques offer promising avenues for enhancing animal breeding programs by leveraging genomic and phenotypic data to predict valuable traits accurately. In this study, we evaluated seven machine learning algorithms viz., K-nearest Network (KNN), Multiple Linear Regression (MLR), Bayesian Regression (BR), Support Vector Machine (SVM), Artificial Neural Network (ANN), Random Forest (RF), and Gradient Boosting Machine (GBM) for predicting genetic merits of Harnali sheep using pedigree and phenotypic information. The dataset comprised records of 2036 Harnali lambs spanning from 1998 to 2021, with predictors including pedigree records, birth year, sex, weight at lambing, birth weight, weaning weight, average daily gain, and six-month body weight. Breeding values for six-month body weight were estimated using the restricted maximum likelihood method under Wombat software. Machine learning algorithms were trained and tested on a 75% train set and 25% test set, respectively. The algorithms were compared based on goodness of fit criteria including coefficient of determination (R), root mean square error (RMSE), mean absolute error (MAE), Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) and bias. The results revealed GBM as the top-performing model, achieving an R value of 0.64 and demonstrating superior predictive accuracy (r = 0.80) with lower values of RMSE, MAE, AIC, BIC and bias. This model consistently outperformed others, showcasing its effectiveness in accurately predicting breeding values for Harnali sheep. The study underscores the potential of machine learning algorithms especially GBM in optimizing breeding programs and accelerating genetic progress in Harnali sheep.
机器学习技术为改进动物育种计划提供了有前景的途径，通过利用基因组和表型数据来准确预测有价值的性状。在本研究中，我们评估了七种机器学习算法，即K近邻网络（KNN）、多元线性回归（MLR）、贝叶斯回归（BR）、支持向量机（SVM）、人工神经网络（ANN）、随机森林（RF）和梯度提升机（GBM），以使用系谱和表型信息预测哈纳利羊的遗传价值。数据集包含1998年至2021年期间2036只哈纳利羔羊的记录，预测变量包括系谱记录、出生年份、性别、产羔时体重、出生体重、断奶体重、平均日增重和六个月体重。使用Wombat软件中的限制最大似然法估计六个月体重的育种值。机器学习算法分别在75%的训练集和25%的测试集上进行训练和测试。根据拟合优度标准对算法进行比较，包括决定系数（R）、均方根误差（RMSE）、平均绝对误差（MAE）、赤池信息准则（AIC）和贝叶斯信息准则（BIC）以及偏差。结果显示GBM是表现最佳的模型，R值为0.64，并且在RMSE、MAE、AIC、BIC和偏差值较低的情况下表现出卓越的预测准确性（r = 0.80）。该模型始终优于其他模型，展示了其在准确预测哈纳利羊育种值方面的有效性。该研究强调了机器学习算法尤其是GBM在优化哈纳利羊育种计划和加速遗传进展方面的潜力。
