基于有监督的多视角变分图自编码器的协同致死基因预测算法

协同致死关系是开发靶向抗癌药物的重要方法之一,通过计算方法预测协同致死基因可以为生物学上的研究提供目标指导,从而提高研究效率并降低实验成本.针对协同致死预测问题,提出了一种通用的、多视角变分图自编码器框架,引入了已知的协同致死关系作为监督信号,同时对局部的单视角数据和全局的多视角协同致死关系重构进行监督训练,在细粒度下获取每个视角中和协同致死相关的基因隐藏表示,最后将多视角的重构图融合在一起进行协同致死预测.在SynLethDB数据集上的实验结果表明方法的有效性.

Synthetic lethality prediction via supervised multi-view variational graph auto-encoder

Synthetic lethality is an important concept for the development of targeted anti-cancer drugs. Predicting synthetic lethality genes through computational methods can provide target guidance for biological research, thus improving research efficiency and reducing experimental costs. This paper proposed a general, multi-view variational graph auto-encoder framework to solve the synthetic lethality prediction problem. The algorithm introduced known synthetic lethality interactions as a supervise signal and performed supervised training on both local single-view data and global multi-view synthetic lethality graph reconstructions. Through supervised training, the algorithm found the latent representation associated with synthetic lethality in each view at a fine-grained level and fused them together for synthetic lethality prediction. Experimental results on the SynLethDB dataset demonstrate the effectiveness of the method.