GlycaNER is an automated tool that detects and normalizes glycan structure terms
(GSTs) in biomedical literature. A BioBERT model, fine-tuned on 5,200 sentences
from 601 PubMed abstracts (4,725 expert-reviewed terms), finds candidate terms.
Post-processing then filters non-glyco terms, resolves conjunctions and
parentheses, and assigns each term a type: GST, glycoconjugate (GCT), sugar,
sequence, composition or generic. Finally, each term is normalized against the
Glycan Structure Dictionary (GSD)
using the glycan structural components in Dictionary Components/.
docker pull shovan5795/glycaner:latestdocker run --rm -v "$PWD/pmids.txt:/data/input/pmids.txt:ro" -v "$PWD/glycaner_output:/data/output" shovan5795/glycaner:latestInput: pmids.txt, a text file with one PubMed ID per line
(see examples/pmids_example.txt). Abstracts are
downloaded from PubTator, so the run needs internet access.
Output: glycaner_output/final/ contains the detected glyco-terms for each
sentence, with their type and normalized GSD match (9c_normalized_form.xlsx),
plus an HTML view of the abstracts with the detected terms highlighted. See
examples/output_45_abstracts for the output of
the 45 example PMIDs.
| Path | Content |
|---|---|
Scripts/BERT_Detection/ |
PubTator download, sentence splitting and BioBERT term detection |
Scripts/Post_Processing/ |
term boundary correction, classification and GSD normalization (run_pipeline.sh gives the order) |
model/ |
fine-tuned BioBERT model (Git LFS) |
Dictionary Components/ |
glycan structural components used for extension and normalization |
Data/ |
annotated training dataset and evaluation set |
Outputs/ |
GlycaNER output for 10,000 PubMed abstracts |
Dockerfile, requirements.txt |
image definition and Python dependencies |
If you use GlycaNER, please cite:
Bhowmik S, Ranzinger R, Au Yeung CCH, Vora J, Bhuiyan U, Tiemeyer M, Mazumder R, Vijay-Shanker K. GlycaNER: An Automated Glycan Structure Term Detection and Normalization Tool in Biomedical Literature. Manuscript in review.