DEFINITION The largest source of biomedical knowledge is the published literature, where results of experimental studies are reported in natural language. Published literature is hard to query, integrate computationally or to reason over. The task of reading published papers (or other forms of experimental results such as pharmacogenomics datasets) and distilling them down into structured knowledge that can be stored in databases as well as knowledgebases is called curation. The statements comprising the structured knowledge are called annotations. The level of structure in annotation statements can vary from loose declarations of “associations“ between concepts (such as associating a paper with the concept ‘colon cancer’) to statements that declare a precisely defined relationship between concepts with explicit semantics. There is an inherent tradeoff between the level of detail of the structured annotations and the time and effort required to create them. Curation to create highly structured and computable annotations requires PhD level individuals to curate the literature. In the molecular biology research community, this task is performed primarily by curators employed by genome databases such as the saccharomyces genome database1. In the biomedical research community this task is performed by curators employed by community portals such as AlzForum for Alzheimer’s research [2] and PharmGKB for pharmacogenomics [3]. In the medical community such curation is still an ignored task, with some groups, such as RCTBank [4], pioneering the effort to curate clinical trial reports.
HISTORICAL BACKGROUND In the biomedical domain, curation began with the formation of cDNA, EST and gene sequence databases such as GenBank. Initially, curation was restricted to the task of assigning a functional annotation (usually in free text) to a sequence being submitted to GenBank. Scientists performing the experiments and submitting the data performed the task on their own. With the rise in the amount of sequence data and subsequently data on the function, structure and cellular locations of gene products along with the formation of communities of researchers around specific model organisms, the task of curation gradually became centralized in the role of a curator at model organism databases. Interaction amongst the curators and leading scientists led to the creation of projects such as the gene ontology project5 in 1998, which led to a systematic basis for creating annotations about the molecular function, biological process and cellular locations of gene products. In subsequent years, user groups formed around other kinds of data, such as microarray gene expression data, resulting in the creation of information models for structuring the metadata pertaining to high throughput experiments. Individual research groups, such as Ecocyc, have already maintained a high level of curation effort, particularly for Information about biological pathways; although it was the success of the gene ontology project that resulted in the widespread appreciation for the need of curated content. With the continued rise in the amount and diversity of biomedical data, the need for curation continues to increase; both in terms of the number of man-hours required and in the level detail desired in the resulting annotations.
