变分图自编码器是图嵌入研究中重要的深度学习模型,但存在着先验正态分布缺陷、训练过程中容易出现后验塌陷等问题.本文从建立云概念空间与隐空间的映射关系入手,引入云模型数字特征对网络中的节点进行不确定性概念表示,设计了一种基于...变分图自编码器是图嵌入研究中重要的深度学习模型,但存在着先验正态分布缺陷、训练过程中容易出现后验塌陷等问题.本文从建立云概念空间与隐空间的映射关系入手,引入云模型数字特征对网络中的节点进行不确定性概念表示,设计了一种基于多维云模型的变分图自编码器(Variational Graph Autoencoder based on Multidimensional Cloud Model,MCM-VGAE).该模型实现了隐空间的多维云概念嵌入及相应的漂移性损失度量,将先验分布扩展为泛正态分布,利用多维正向云发生器及云包络带修正采样算法实现了重参数化过程,有效缓解了后验塌陷现象.在应用效果上,模型在多类型数据集上的链路预测、节点聚类、图嵌入可视化实验表现均优于基准模型,进一步说明了方法的普适有效性.展开更多
The function of the 3 040 bp sequence at the upstream translation starting site (ATG) of the ZAG2 gene, isolated from the maize genome, was studied. The sequence analysis showed that the sequence contained a typical...The function of the 3 040 bp sequence at the upstream translation starting site (ATG) of the ZAG2 gene, isolated from the maize genome, was studied. The sequence analysis showed that the sequence contained a typical class C MADS-box gene regulatory element. The 5′ UTR region of the gene contains a 1 299-bp intron that might have important regulatory functions. To study the sequence function, deletion derivatives of promoter-reporter (uidA) gene fusions were generated and transformed into tobaccos. The GUS staining and fluorescence quantification results showed that the GUS activity was detected only in the third and fourth whorl floral organs of the transgenic tobaccos under driving the promoter including the first intron, while detected in all the organs and was stronger under driving the promoter without the first intron. However, the GUS activity was just detected in one whorl of the fourth or third floral organs under driving of the 35S promoter. These results suggested that the first intron of the ZAG2 gene contains functional regulatory elements, which turned out to be important for gene expression in the heterologous systems. Moreover, the GUS activity was decreased when the reporter gene driven by the promoters with 5′-deletions, respectively, from -1 606 to -951 and -951 to -426 nts, which indicates that positive regulatory elements are present in these two sequence stretches.展开更多
文摘变分图自编码器是图嵌入研究中重要的深度学习模型,但存在着先验正态分布缺陷、训练过程中容易出现后验塌陷等问题.本文从建立云概念空间与隐空间的映射关系入手,引入云模型数字特征对网络中的节点进行不确定性概念表示,设计了一种基于多维云模型的变分图自编码器(Variational Graph Autoencoder based on Multidimensional Cloud Model,MCM-VGAE).该模型实现了隐空间的多维云概念嵌入及相应的漂移性损失度量,将先验分布扩展为泛正态分布,利用多维正向云发生器及云包络带修正采样算法实现了重参数化过程,有效缓解了后验塌陷现象.在应用效果上,模型在多类型数据集上的链路预测、节点聚类、图嵌入可视化实验表现均优于基准模型,进一步说明了方法的普适有效性.
基金supported by the National Major Project for Transgenic Organism Breeding, China (2011ZX08003-001)
文摘The function of the 3 040 bp sequence at the upstream translation starting site (ATG) of the ZAG2 gene, isolated from the maize genome, was studied. The sequence analysis showed that the sequence contained a typical class C MADS-box gene regulatory element. The 5′ UTR region of the gene contains a 1 299-bp intron that might have important regulatory functions. To study the sequence function, deletion derivatives of promoter-reporter (uidA) gene fusions were generated and transformed into tobaccos. The GUS staining and fluorescence quantification results showed that the GUS activity was detected only in the third and fourth whorl floral organs of the transgenic tobaccos under driving the promoter including the first intron, while detected in all the organs and was stronger under driving the promoter without the first intron. However, the GUS activity was just detected in one whorl of the fourth or third floral organs under driving of the 35S promoter. These results suggested that the first intron of the ZAG2 gene contains functional regulatory elements, which turned out to be important for gene expression in the heterologous systems. Moreover, the GUS activity was decreased when the reporter gene driven by the promoters with 5′-deletions, respectively, from -1 606 to -951 and -951 to -426 nts, which indicates that positive regulatory elements are present in these two sequence stretches.