KeDaCoYa/MKG-GC

Introduction

Paper:

MKG-GC: a multi-task learning-based knowledge graph framework with personalized application to gastric cancer

The project includes code for entity extraction, entity normalization, relational classification, and knowledge embedding.

Dataset

Download link for the dataset：

• baidu drive: link code: ezvp
• google drive：link

Note： Due to the limitations of the agreement between UMLS and DrugBank, we are unable to provide corresponding data

entity extraction dataformat

BIO format for example:

gastric B-Disease
cancer I-Disease
is O
a O
malignant B-Disease
disease I-Disease
. O

entity normalization dataformat

Each medical entity corresponds to an ID, and multiple medical entity names have the same ID For Example:

gastric cancer|D013274
stomach neoplasm|D013274
cancer|D009369

relation classification dataformat

Two entities of each piece of data are surrounded by special tokens Text|entity1|entity2|entity_type1|entity_type2

We found that the expression of all these genes, including thymidylate synthase (TS),  [s2] dihydrofolate reductase [e2]  (DHFR), ribonucleotide reductase (PNR), proliferating cell nuclear antigen (PCNA),  [s1] histone [e1]  H1, histone H2A + 2B, histone H3, and histone H4, was induced to high levels in young IMR-90 cells but not in old IMR-90 cells.	histone	dihydrofolate reductase	protein	protein	

Run

You need to download biobert first: https://github.com/dmis-lab/biobert

entity extraction

cd entity_extraction
bash multi_run.sh

entity extraction

cd entity_extraction
bash multi_run.sh

entity normalization

cd entity_normalization
bash multi_train.sh

relation classification

cd relation_classification
bash multi_run.sh

requirements

python>=3.8
transformer == 3.0.0
pytorch >= 1.7

Contact Information

For help or issues using MKG-GC, please submit a GitHub issue.