Automated extraction of information from the literature on chemical-CYP3A4 interactions

A text mining system is presented for automatically extracting information from the literature on chemical-CYP3A4 interactions (i.e., substrate, induction, inhibition). The system identifies chemicals and CYP3A4 forms according to a combination of name dictionaries and context features. In addition, it transforms sentences into multiple simple clauses each containing a single event and extracts information on chemical-CYP3A4 interactions using a simple but effective pattern matching method based on the order of three keywords (chemicals, CYP3A4, key verbs). Using this system, 2990 relations including 2700 identified interactions with CYP3A4 for 600 chemicals were extracted from a corpus of 2900 PubMed abstracts. In an evaluation test using 100 randomly selected abstracts, it achieved 87.4% recall and 92.3% precision for identification of the chemical name and 85.2% recall and 92.0% precision for the extraction of chemical-CYP3A4 interactions, respectively. This system will be applicable to interactions of chemicals with any functional proteins, such as enzymes and transporters, simply by changing the list of key verbs.