Dataset construction

A comprehensive MeSH dataset is fundamental to MeSH tools. However, the MeSH dataset used by most popular MeSH tools contains only MeSH Main Headings (MHs), a component of MeSH descriptor records, but it contains no SCRs. To construct a comprehensive MeSH dataset, we extracted MeSH information, including MHs, SCRs, and their relations, from Unified Medical Language System (UMLS, 2018AA version) which is a large biomedical thesaurus integrating nearly 200 vocabularies including MeSH [14].

The multiple-to-one relationship between MeSH-synonymous UMLS concepts and MeSH MHs was curated from the table MRSAT in UMLS. For example, the MeSH MH “Alzheimer Disease” (D000544) includes seven MeSH concepts, each of which corresponds to several MeSH entry terms and a UMLS concept (Supplementary Table 1). In our dataset, we included the MeSH MHs and related UMLS concepts, while we excluded the MeSH concept and MeSH entry term information. Moreover, we curated the most useful “parent” and “child” relationship between MeSH MHs from the table MRREL in UMLS.

The one-to-one relationship between MeSH-synonymous UMLS concepts and SCRs was curated from the table MRSAT in UMLS. In our dataset, we included the SCRs and its corresponding UMLS concepts, as well as the “narrower” and “broader” relationship between SCRs and MeSH MHs curated from the table MRREL in UMLS.

The qualifier records and other MeSH descriptor records except MeSH MHs were not included in our dataset. In the study, we used “MeSH term” to refer to MeSH MH or SCR.