Skip to main content

Table 2 Prediction of drug-target like documents from PubMed articles. The third column shows the number of documents that contain either drug or protein entities as identified by PubTator. In contrast, the fourth column indicates the number of documents that contain both drug and protein entities

From: Using BERT to identify drug-target interactions from whole PubMed

BERT model

Predicted as drug-target articles

Articles containing drugs or proteins on PubTator

Articles containing both drugs and proteins on PubTator

BERT

688,206

682,150

342,902

SciBERT

594,999

589,999

321,831

BioBERT

636,091

630,132

340,638

BioMed-RoBERTa

725,748

720,030

385,015

BlueBERT

570,284

564,220

297,834

Majority voting

597,844

592,789

316,794

  1. Bold value indicates the top result for a dataset