NEUROINFORMATICS ORIGINAL RESEARCH ARTICLE

We hypothesize that many of the methods and approaches devel-oped for extraction of information about genes can be applied to extraction of information about brain areas. This is an attrac-tive approach because many of the challenges in analyzing text for information about genes are also faced in trying to mine informa-tion about brain regions. These challenges include abbreviations, synonyms, lexical variation and ambiguity. For example, the gene “carbonic anhydrase 1 ” has synonyms including “carbonate dehy-dratase I”, “Car1”, and “CA-I”. Its offi cial symbol, CA1, is ambiguious in that it also matches a drug (the abbreviated form of coumermy-cin A1) and a brain region (the CA1 fi eld of the hippocampus). Similarly brain regions have a variety of names and abbreviations, and can be confused with other types of entities. Approaches have been developed for addressing these problems for genes, so it seems reasonable to expect that the lessons learned will apply at least partly to other domains. However, before these approaches can be applied to brain regions, a “gold standard ” corpus is needed. Such a corpus is needed both as training data for algorithms and for evaluation of methods. To our knowledge, no such resource exists for neuroscience text mining. Past efforts in neuroscience text mining provided limited ability to retrieve brain region mentions, by looking for exact matches of brain region names from small lists (Crasto et al., 2003, 2007; Muller et al., 2008). This limits the recall to a small number of (usually broad or large) brain regions. The most extensive effort is “Textpresso fo