基于半监督符号图神经网络聚类的药物社区发现

Drug community discovery based on semi-supervised signed graph neural network clustering

药物社区反映了多组药物之间的互作用关系,其内涵整个药物网络的结构与功能,考虑药物社区发现问题有助于推动新药物的研发。大多数图神经方法仅局限于挖掘单一药物之间的互作用,而忽略了多种药物共同作用对药理的强化或抑制。通过构建带符号的药物互作用网络,将药物社区发现问题转化为符号网络中的聚类问题,提出的半监督符号神经网络聚类方法可以很好地弥补这一空缺。交叉验证实验表明,所提出的聚类方法具有较好的性能,在药物分析中具有潜在应用价值。

The drug community reflects the interaction relationship between multiple groups of drugs,which encompasses the structure and function of the entire drug network.Considering the discovery of problems in the drug community can help promote the development of new drugs.Most graph neural methods are limited to exploring the interactions between single drugs,while neglecting the reinforcement or inhibition of pharmacology by the combined effects of multiple drugs.By constructing a signed drug interaction network,the problem of drug community discovery can be transformed into a clustering problem in the symbolic network.A semi supervised symbolic neural network clustering method can effectively fill this gap.Cross validation experiments have shown that the proposed clustering method has good performance and potential application value in drug analysis.