基于卷积神经网络的仿刺参 （Apostichopus japonicus）疣足全基因组选择初步研究

A Preliminary Study on Genome-wide Selection of Sea Cucumber (Apostichopus japonicus) Tubercles using Convolutional Neural Networks

中文关键词:  仿刺参  全基因组选择  卷积神经网络  疣足数量

英文关键词:Apostichopus japonicus  genomic selection  CNN  papillae number

中文摘要:

      仿刺参 （Apostichopus japonicus） 是山东省和辽宁省的支柱水产种类之一。全基因组选择 （Genomic Seletion，GS） 分子育种以高效率、高准确率等优势，将农业生物育种推向了新的发展阶段。目前，传统家系育种仍是刺参良种选育的主要手段，发展适用于刺参的全基因组选择技术是刺参良种选育的未来重点发展方向，具有重要研究价值。卷积深层神经网络 （Convolutional Neural Network，CNN） 已经应用在基因互作效应的估计中，解决全基因组选择主流 GBLUP 方法无法估计基因互作的非加性效应的问题。本研究以仿刺参的重要经济性状疣足为表型，使用 52 271 个 SNP 基因型，探究了 GBLUP 和 CNN 育种值估计的差异。结果表明，在仿刺参疣足数量的遗传力为 0.611的情况下，相比于 GBLUP 方法的预测准确率 （0.563 63），CNN 方法的预测准确率 （0.834 85） 更高, 表明基于 CNN 学习 SNP 之间潜在的基因互作信息，提高了育种值估计的准确率。本研究首次将 CNN 用于仿刺参疣足性状的全基因组选择中，研究结果为仿刺参生长性状的分子育种提供了理论支撑。

英文摘要:

      Apostichopus japonicus, commonly known as the sea cucumber, is widely cultivated in China due to its high nutritional value. Recently, molecular breeding has gradually replaced traditional breeding methods due to its high efficiency and accuracy. However, in the breeding of A. japonicus, the accuracy of estimates based on common breeding models remains relatively low, failing to meet the demands of actual production. The number of papillae in A. japonicus is an important economic trait and a key indicator for breeding improvement. Deep learning methods can be applied in molecular breeding to address limitations of the GBLUP method, which considers only additive genetic effects. This study used the papillae count of 215 A. japonicus as the phenotype and a dataset of 52 271 SNPs to compare the breeding value estimation accuracy between GBLUP and CNN. The results showed that when the heritability of the number of papillae was 0.611, the CNN model achieved a prediction accuracy of 0.834 85, which was higher than the GBLUP method (0.563 63). These findings suggest that CNN can capture potential epistatic interactions among SNP, effectively overcoming the limitations of GBLUP and enhancing precision of breeding value estimation. This study is the first to apply CNN to whole-genome selection breeding in A. japonicus, and the results not only promote the advancement of molecular breeding for A. japonicas, but also provide theoretical support for its artificial cultivation.