Skip to main content
. 2018 Jan 17;19:15. doi: 10.1186/s12859-018-2021-9

Fig. 1.

Fig. 1

Most related protein and localizations closed to each other. Repetitions of relationships were collapsed at the document level after normalizing the entities: proteins to UniProtKB and localizations to GO. In the LocTextCorpus, the majority of unique relations were annotated between entities occurring in the same sentence (distance 0 = D0; 66% of all relations) or in adjacent sentences (dist. 1 = D1; 15%). Combined, D0+D1 accounted for 81% of the relations. Removing repetitions when considering the GO hierarchy (children identifiers are more exact than their parents), D0+D1 accounted for 89% of all unique relationships