Mining connections between chemicals, proteins, and diseases extracted from Medline annotations

Abstract

The biomedical literature is an important source of information about the biological activity and effects of chemicals. We present an application that extracts terms indicating biological activity of chemicals from Medline records, associates them with chemical name and stores the terms in a repository called ChemoText. We describe the construction of ChemoText and then demonstrate its utility in drug research by employing Swanson's ABC discovery paradigm. We reproduce Swanson's discovery of a connection between magnesium and migraine in a novel approach that uses only proteins as the intermediate B terms. We validate our methods by using a cutoff date and evaluate them by calculating precision and recall. In addition to magnesium, we have identified valproic acid and nitric oxide as chemicals which developed links to migraine. We hypothesize, based on protein annotations, that zinc and retinoic acid may play a role in migraine. The ChemoText repository has promise as a data source for drug discovery.

Figure 1

Swanson’s ABC Paradigm

Figure 2

Swanson’s ABC using ChemoText

Figure 3. Medline processing into data tables

The top part of the figure shows selected MeSH annotations in the Medline record for PubMed ID 16640785. The bottom of the figure shows the database entries in ChemoText that result from the processing of this Medline record.

Figure 4

Schematic view of ChemoText

Figure 5

Bar chart showing percentages by protein count

Figure 6

Bar chart showing percentages by protein count for chemicals with 10 or fewer associated proteins