Fig 1.
The BEST system consists of two main parts: Indexing and Searching. (1) “Indexing” represents the indexing subsystem of BEST. For every document, BEST extracts all biomedical entities (1-a) and makes a paired posting (1-b). The basic structure of BEST’s index is similar to that of the inverted index of conventional search engines. However, BEST uses a different indexing unit, paired posting, which is a pair of a document ID and a list of entities that appear in the document. (2) “Searching” represents the search subsystem of BEST. All retrieved paired postings are aggregated to rank the entities (2-a). Ranking scores are computed using four subcomponents described in “Searching and Scoring” in the Methods section (2-b).
Fig 2.
Users can pose queries in the query window (a) and select result entity types either by using the drop down box (b) or by clicking on the entity-type filter tab (c). BEST returns a list of entities that are relevant to a user’s query. For each entity in the list, BEST shows a description (d), an interaction network (e), enriched GO terms (f) of the entity, and top 3 abstracts (g) in which the query terms and the entity co-occur.
Table 1.
Search result of query "imatinib resistance ABL1" with type filter "mutations."
Fig 3.
BEST’s result of "MAP2K1" with type filter "genes."
Table 2.
Top 10 drugs returned for query "chronic myeloid leukemia."
Table 3.
Accuracy and response time comparison of Best, PolySearch2, and FACTA+.
Fig 4.
Recency evaluation of BEST using "(chronic myeloid leukemia) AND (year:[*—YYYY])" with result type filter “drug.”
Table 4.
Search results of drugs when more weight is given to the recency factor.
Table 5.
Source databases for BEST dictionary.