|
Paper | Relation type | Knowledge base | Corpora | NER results | RE results |
|
[12] | Protein-residue | Protein Data Bank (PDB) [25] | PubMed abstracts | Evaluated on 3 gold corpora only for amino acid/mutation entities: Nagel et al. F-measure = 93.28%/mutation finder: development (F-measure = 89.32%) and test corpora (F-measure:88.04%) LEAP-FS corpus: F-measure = 86.56% | 0.84 F-measure (silver corpus) 0.79 F-measure (gold corpus) |
|
[13] | Drug-drug protein-protein | IntAct database [26], KUPS database [27], DrugBank [28] | The five corpora of Pyysalo et al. [29]. The corpus of Segura-Bedmar et al. [30] | Not mentioned | Drug-drug (DDI) F-score = 61.19 PPI F-score = 78.0 on LLL corpus |
|
[14] | Gene-brain regions | UMLS Semantic Network [31] | 10,000 randomly selected full-text articles from Elsevier Neuroscience corpus | F1 = 0.8 (for 300 manually examined examples) | F1-score = 0.468, recall = 0.459, precision = 0.477 (for 259 manually labelled sentence out of 30,000) |
|
[15] | Protein-location | UniProtKB (Swiss-Prot) [32] | 43,000 full-text articles from the Journal of Biological Chemistry | Not mentioned | F1 = 0.61, R = 0.49, P = 0.81 (sentence level) accuracy = 0.57 (RL instance level) |
|
[16] | microRNA-gene | TransmiR database (nonhuman entries) [33] | IBRel-miRNA corpus | Evaluated on 3 corpora: Bagewadi corpus [34] (F = 0.919 miRNA/F = 0.677 gene), miRTex [35] (F = 0.941 miRNA/F = 0.795 genes), and TransmiR (F = 0.687 miRNA/F = 0.361 genes) | Evaluated on 3 corpora: Bagewadi corpus (F = 0.532), miRTex (F = 0.383), and TransmiR (F = 0.413) |
|
[17] | Related symptoms, related diseases, related examination, complications, and related treatment | Not mentioned | Medical websites | Not mentioned | Accuracy = 91.87%, recall = 91.58%, F1-score = 0.8908 |
|
[18] | Gene-drug | Gene Drug Knowledge Database (GDKD) [36] | Biomedical literature from PubMed Central | Not mentioned | Automatic evaluation best average test accuracy in fivefold cross-validation (single sentence: 88, cross sentence: 87.5) manual evaluation (precision = 71 for single sentence and 61 for cross sentence) |
|
[19] | n-arity relations: Treats, ReducesRisk, Causes, Diagnoses | 474 seed facts from online medical portals uptodate.com, drugs.com | Encyclopaedic articles and PubMed scientific publications | Not mentioned | Treats avg. precision: 0.86, ReducesRisk avg. P: 0.82, Causes avg. P: 0.80, and Diagnoses avg. P: 0.89 |
|
[20] | Protein-protein, protein-location | IntAct database, UniProt database | Medline, literature found in IntAct database | Not mentioned | PPI (PCNN F-score = 56.8 BiLSTM F-score = 50.4) PLOC (PCNN F-score = 54.5 BiLSTM F-score = 60.4) |
|
[21] | Chemical-disease | Comparative Toxicogenomics Database (CTD Database) [37] | PubMed abstracts | Not mentioned | Intrasentence level: best F-score = 60.8; intersentence level: best F-score = 22.8 |
[22] | Binary treatment relation | UMLS database, SemMedDB [38] | PubMed abstracts for which there exist both the therapeutic use and the therapy medical subject headings (MeSH) subheadings | Not mentioned | PR-AUC: logistic regression:82.86 BiLSTM:81.18 BiLSTM-NLL:81.38 |
|
[23] | Human disease-gene, tissue-gene, and protein-protein in different species | Genetics Home Reference (GHR) [39], UniProtKB, KEGG maps [40], STRING [41] | PubMed, full-text articles from PMC in BioC XML format [42] | Not mentioned | Adjusted area under the precision-recall curve (AUPRC): disease-gene: 0.86/tissue-gene: 0.19 |
|