摘要
采用分子压缩的方法提出了一种压缩条件流模型(CompMF),用来解决分子生成问题。实验结果表明,模型在分子生成任务中的表现优异,生成的分子在保证有效性的同时,新颖性和唯一性均有所提高,同时模型也降低了分子数据的维度。由此得出结论,将分子压缩技术与流模型相结合在分子生成任务中有着显著的意义。
A compressed conditional flow model(CompMF)by adopting the molecular compression method is proposed to solve the problem of molecular generation.The result of experiments show that the model performs a better performance.The novelty and uniqueness of generated molecules have been improved while ensuring their validity.And the model also reduces the dimension of molecular data.The conclusion is that the combination of molecular compression technology and flow model has a significance in the molecular generation mission.
作者
刘晨阳
刘勇
惠丽
LIU Chen-Yang;LIU Yong;HUI Li(College of Computer Science and Technology,Heilongjiang University,Harbin 150080,China)
出处
《黑龙江大学工程学报》
2022年第4期77-83,共7页
Journal of Engineering of Heilongjiang University
基金
国家自然科学基金面上项目(61972135)
黑龙江省自然科学院基金项目(LH2020F043)。
关键词
图卷积
流模型
图生成
分子生成
分子压缩
深度学习
graph convolution
flow model
graph generation
molecular generation
molecular compression
deep learning
