Building Large-Scale Gene-Disease Association Datasets for Biomedical Relation Extraction

This repository contains the GDAb and GDAt datasets. GDAb and GDAt are large-scale, distantly supervised, and manually enhanced datasets for Gene-Disease Association (GDA) extraction. Each dataset consists of three text files, corresponding to train, validation, and test sets, plus an additional JSON file containing the mapping between relation names and IDs. Each record in train, validation, or test files corresponds to a single GDA extracted from a sentence. Records are represented as JSON objects with the following structure: text: sentence from which the GDA was extracted. relation: relation name associated to the given GDA. h: JSON object representing the gene entity, composed of: id: UMLS CUI associated to the gene entity. name: UMLS preferred name associated to the gene entity. pos: list consisting of starting position and length of the gene mention within text. t: JSON object representing the disease entity, composed of: id: UMLS CUI associated to the disease entity. name: UMLS preferred name associated to the disease entity. pos: list consisting of starting position and length of the disease mention within text. Both datasets contain over 2,500,000 sentences and 500,000 bags. The zip file consists of two folders, GDAb and GDAt,  containing the files corresponding to the two datasets, respectively.