Figure 1

The system architecture of MeInfoText 2.0. DNA methylation-related abstracts were collected from PubMed and processed by sentence splitting and expansion. Then, our methylation and gene mention tagger annotated the collected abstracts with methylation terms and gene names, after which the relations between two entities were extracted by two trained maximum entropy (ME) models. The first model determined the gene-methylation (G-M) relation. All positive sentences were annotated with the type of cancer by using a pattern-based cancer type tagger. The second ME model extracted gene-cancer relations. The gene information, methylation statistics and associations ranked in descending order of probability are stored in MeInfoText 2.0. Users can query the database via the web interface using gene names or cancer types, or a combination of both.