知识图谱驱动的人类蛋白质组知识注释与知识探索研究

Knowledge Graph Powered Human Proteome Knowledge Annotation and Knowledge Exploration Study

蛋白质组知识注释能够帮助科研人员从已有知识中凝练出重要的科学假说,传统的蛋白质组知识注释知识单一且多为简单的知识检索和罗列汇总,知识的综合性和系统性不足。为此,本研究展开知识图谱驱动的人类蛋白质组知识注释与知识探索研究。综合13种生物医学本体和数据库资源,依托图数据库Neo4j构建知识图谱(BMKG)管理生物医学知识;结合先验知识和中心度等图算法,从多层面多角度设计元路径,制定知识注释方案;实现相似度计算和链接预测算法,执行node2vec图嵌入,为辅助知识探索分析做准备。本研究所得BMKG图谱融合了13种生物医学知识资源,共9种2508348个节点,20种25362594条关系。以肾细胞癌蛋白质组知识注释为例,BMKG的知识注释方案能够从多层面多角度展开注释,实现对其关联的通路、药物和表型等知识的归纳提炼。此外,基于BMKG能够展开药物-疾病关联预测等知识探索研究,疾病知识的聚类结果与Mondo本体结构有很好的一致性。本研究建设了网站提供在线服务,网址:http://bmkg.bmicc.org,包括三大应用模块:知识检索、知识注释和知识分析。综上,本研究表明知识图谱能够为人类蛋白质组知识注释研究提供新的见解。

Proteome knowledge annotation facilitates the derivation of scientific hypotheses from existing knowledge.However,traditional annotation approaches are often not comprehensive and lack systemic integration,being limited to knowledge retrieval and aggregation.In this paper,a novel method involving knowledge graphs is proposed to integrate biomedical knowledge from 13 biomedical ontologies and databases.The knowledge graph,Biomedical Knowledge Graph(BMKG),was constructed with the graph database Neo4j.Metapaths were designed to create knowledge annotation schemes which incorporated prior knowledge with graph algorithms such as centrality measures.By leveraging similarity calculations,link prediction algorithms,and node2vec graph embedding,knowledge exploration analysis was facilitated.BMKG encompasses 2508348 nodes of 9 types and 25362594 relationships of 20 types.The BMKG knowledge annotation scheme facilitates diverse perspectives and multi-level annotation,which is demonstrated by its application to renal cell carcinoma tissue proteome data in annotating various biological aspects comprehensively,such as pathways,drugs,and phenotypes.Additionally,BMKG supports knowledge exploration studies,such as drug-disease association prediction,and the clustering of disease knowledge exhibits strong concordance with the Mondo ontology structure.Moreover,an online platform(http://bmkg.bmicc.org)has been established,with three analysis modules:knowledge retrieval,knowledge annotation,and knowledge analysis.Collectively,this study demonstrates the potential of knowledge graph approaches to enhance human proteome knowledge annotation and knowledge exploration.