Zhao Wei, Lai Xueshuang, Liu Dengying, Zhang Zhenyang, Ma Peipei, Wang Qishan, Zhang Zhe, Pan Yuchun
Department of Animal Science, School of Agriculture and Biology, Shanghai Jiao Tong University, Shanghai, China.
Department of Animal Science, College of Animal Science, Zhejiang University, Hangzhou, China.
Front Genet. 2020 Dec 3;11:598318. doi: 10.3389/fgene.2020.598318. eCollection 2020.
Genomic prediction (GP) has revolutionized animal and plant breeding. However, better statistical models that can improve the accuracy of GP are required. For this reason, in this study, we explored the genomic-based prediction performance of a popular machine learning method, the Support Vector Machine (SVM) model. We selected the most suitable kernel function and hyperparameters for the SVM model in eight published genomic data sets on pigs and maize. Next, we compared the SVM model with RBF and the linear kernel functions to the two most commonly used genome-enabled prediction models (GBLUP and BayesR) in terms of prediction accuracy, time, and the memory used. The results showed that the SVM model had the best prediction performance in two of the eight data sets, but in general, the predictions of both models were similar. In terms of time, the SVM model was better than BayesR but worse than GBLUP. In terms of memory, the SVM model was better than GBLUP and worse than BayesR in pig data but the same with BayesR in maize data. According to the results, SVM is a competitive method in animal and plant breeding, and there is no universal prediction model.
基因组预测（GP）彻底改变了动植物育种。然而，需要能提高GP准确性的更好的统计模型。因此，在本研究中，我们探索了一种流行的机器学习方法——支持向量机（SVM）模型基于基因组的预测性能。我们在八个已发表的猪和玉米基因组数据集中为SVM模型选择了最合适的核函数和超参数。接下来，我们在预测准确性、时间和内存使用方面，将具有径向基函数（RBF）和线性核函数的SVM模型与两种最常用的基因组辅助预测模型（GBLUP和BayesR）进行了比较。结果表明，SVM模型在八个数据集中的两个数据集中具有最佳预测性能，但总体而言，两个模型的预测结果相似。在时间方面，SVM模型优于BayesR但不如GBLUP。在内存方面，SVM模型在猪数据中优于GBLUP但不如BayesR，而在玉米数据中与BayesR相同。根据结果，SVM在动植物育种中是一种有竞争力的方法，并且不存在通用的预测模型。
