Using a gene???drug network automatically created from sentence-level co-occurrence data in the full text of scientific articles (1731 documents taken from PharmGKB), Garten et al. [91] compare the performance with that of a network created by manual curation of those articles.