Multi_Level_Biological_Networks

In our work, we collected biological entities and their relations, and put them into these networks. The below description is in our manuscript. As soon as the manuscript is published, we will present the access information in this page. --- The molecule type consisted of genes, and the function type consisted of GO terms. The phenotypic type consisted of diseases that are MeSH terms in disease categories, and the concepts in some UMLS concept types: <em>Congenital Abnormality, Acquired Abnormality, Finding, Pathologic Function, Disease or Syndrome, Mental or Behavioral Dysfunction, Cell or Molecular Dysfunction, Sign or Symptom, Anatomical Abnormality, </em>and<em> Neoplastic Process.</em> Those UMLS concept types contained a considerable number of the MeSH terms in disease categories. 18 types of biological relations were defined with their pre-defined relation types and their original resources. The 18 types are presented in the Supplementary Material (Supplementary Table S2). From the CODA, 13 types of relations for this study were defined by combining the six original resources and the eleven relation types that were pre-defined in the CODA. The six original resources were BioGRID⁴¹, RegNetwork, TRANSFAC⁴², EndoNet⁴³, KEGG, GO, and PhenoGO⁴⁴, which were released until 2016. The six pre-defined types in CODA were: <em>Undirected Link, Directed Link, Positive Increase, Positive Decrease, Negative Increase, </em>and<em> Negative Decrease</em>. Moreover, their reversal types were also made. The <em>Undirected Link</em> means a non-directional association, and the <em>Positive Decrease</em> means that the activity or the amount of a receiver decreases as the activity or the amount of an actor increases¹⁹. From the UMLS (version 2016AA), four types of relations were defined for the present study. They were defined only by their original resources: MedlinePlus⁴⁵, MTHMST⁴⁶, NCI⁴⁷, and OMIM. Although the UMLS has more than 200 resources, only the four resources provided a considerable number of relations among the nodes that we considered. The pre-defined types of relations in the four resources were too many for us to properly categorize them. Therefore, we ignored them when defining the types of relations for this study. The MEDLINE (version 2017) provides the co-occurrence frequency of two MeSH terms. It says how many times two MeSH terms have been attached to the same biomedical literature. Only one type of relations was defined for our research. Among functions or phenotypes, if two nodes had any co-occurrence frequency, we determined there is a co-occurrence relation between them. We tried to separate them according to their frequency. However, the performance, AUROC values, in the application was best when the co-occurrence relations were not separated.本研究中，我们收集了生物实体及其相互关联关系，并将其整合至各类网络结构中。 以下内容来自本研究手稿。待该手稿正式发表后，我们将在本页面公布数据集的获取方式。 本研究的分子类型实体为基因，功能类型实体为基因本体（Gene Ontology，GO）术语。表型类型实体涵盖两类：一是疾病分类下的医学主题词（Medical Subject Headings，MeSH）所收录的疾病，二是统一医学语言系统（Unified Medical Language System，UMLS）中以下概念类型所包含的实体：先天性异常（Congenital Abnormality）、获得性异常（Acquired Abnormality）、异常发现（Finding）、病理功能（Pathologic Function）、疾病或综合征（Disease or Syndrome）、精神或行为功能障碍（Mental or Behavioral Dysfunction）、细胞或分子功能异常（Cell or Molecular Dysfunction）、体征或症状（Sign or Symptom）、解剖结构异常（Anatomical Abnormality）以及肿瘤进程（Neoplastic Process）。上述UMLS概念类型中包含了大量疾病分类下的MeSH术语。 本研究共定义了18种生物关联关系，每种关联均对应其预设的关系类型与原始数据源。这18种关联类型详见补充材料（补充表S2）。 本研究从CODA中选取13种关联关系，其构建方式为结合6种原始数据源与CODA中预设的11种关联类型。上述6种原始数据源分别为BioGRID⁴¹、RegNetwork、TRANSFAC⁴²、EndoNet⁴³、KEGG、GO以及PhenoGO⁴⁴，所有数据源均为2016年及之前发布的版本。CODA中预设的6种关联类型为：无向链接（Undirected Link）、有向链接（Directed Link）、正向增强（Positive Increase）、正向减弱（Positive Decrease）、负向增强（Negative Increase）以及负向减弱（Negative Decrease）。此外，我们还构建了各类关联的反向类型。其中，无向链接指代无方向性的关联关系，正向减弱则指当作用因子的活性或表达量升高时，靶标分子的活性或表达量会随之降低¹⁹。 本研究从统一医学语言系统（UMLS，版本2016AA）中定义了4种关联关系，这些关联仅基于4种原始数据源构建：MedlinePlus⁴⁵、MTHMST⁴⁶、NCI⁴⁷以及OMIM。尽管UMLS拥有超过200种数据源，但仅上述4种数据源可为本研究关注的节点间提供足量的关联关系。由于这4种数据源预设的关联类型过多，难以对其进行合理归类，因此本研究在定义关联类型时未采用这些预设类型。 MEDLINE（版本2017）提供了两个MeSH术语的共现频率数据，即两个MeSH术语被同时标注至同一篇生物医学文献中的次数。本研究仅定义了1种共现关联关系：在功能或表型类型的实体节点中，若两个节点存在共现频率，则判定二者之间存在共现关联。我们曾尝试按照共现频率对该类关联进行分级，但后续应用实验结果显示，不进行分级时的模型性能（受试者工作特征曲线下面积（Area Under the Receiver Operating Characteristic Curve，AUROC）值）最优。