lasigebiotm/biont

BiOnt: Deep Learning using Multiple Biomedical Ontologies for Relation Extraction

BiOnt: Deep Learning using Multiple Biomedical Ontologies for Relation Extraction

To perform relation extraction, our deep learning system, BiOnt, employs four types of biomedical ontologies, namely, the Gene Ontology, the Human Phenotype Ontology, the Human Disease Ontology, and the Chemical Entities of Biological Interest, regarding gene-products, phenotypes, diseases, and chemical compounds, respectively.

Our academic paper which describes BiOnt in detail can be found here.

Downloading Pre-Trained Weights

Available versions of pre-trained weights are as follows:

• DRUG-DRUG
• DRUG-DISEASE
• GENE-PHENOTYPE

The training details are described in our academic paper.

Getting Started

 cd bin/
 git clone [email protected]:lasigeBioTM/DiShIn.git

Use the Dockerfile to set up the rest of the experimental environment or the BiOnt Image available at Docker Hub.

Preparing Data

• $2: type_of_action
• $3: pair_type
• $4: preprocess_what
• $5: input_path

Example:

 python3 src/ontologies_embeddings.py preprocess DRUG-DRUG train corpora/drug_drug/train
 python3 src/ontologies_embeddings.py preprocess DRUG-DRUG test corpora/drug_drug/test

For more options check model.sh.

Train Model

• $2: type_of_action
• $3: pair_type
• $4: model_name
• $6:: channels

Example:

 python3 src/ontologies_embeddings.py train DRUG-DRUG model_name words wordnet concatenation_ancestors common_ancestors

For more options check model.sh.

Predict New Data

• $2: type_of_action
• $3: pair_type
• $4: model_name
• $5: gold_standard OR data_to_test
• $6:: channels

Example:

 python3 src/ontologies_embeddings.py test DRUG-DRUG model_name corpora/drug_drug/test words wordnet concatenation_ancestors common_ancestors

For more options check model.sh.

Reference

• Diana Sousa and Francisco M. Couto. 2020. BiOnt: Deep Learning using Multiple Biomedical Ontologies for Relation Extraction. In Jose J. et al. (eds) Advances in Information Retrieval. ECIR 2020. Lecture Notes in Computer Science, Volume 12036, pages 367-374. Springer, Cham.

biont's Issues

bin/sst-light-0.4/sst

hello,I want to know the location of these files above picture, the directory of "bin" is nothing,can you tell me where them is?

Dataset

Hello,

Thank you for your work. Could you please share the datasets? There are no ontologies in your data as the folders are empty and just have .gitkeep. Your ontologies are OWL/RDF-based?

Assumed bug in preprocess_ids

In ontologies_embeddings.py in the method preprocess_ids, I assume that it's intended to collect the list of all ontology parents/ancestors. But wat it does instead is to take only the last ancestor. I guess correct would be something like:

 idxs.append(id_to_index[d.replace('_', ':')])

May I know where can I get your train/test file?

error while loading saved model

original_keras_version = f.attrs['keras_version'].decode('utf8') AttributeError: 'str' object has no attribute 'decode'

preprocessing addition

capture more entities by allowing overlapping entities to be recognized by spacy

make error for sst-light-0.4

: recipe for target 'sst-light.o' failed
make: *** [sst-light.o] Error 1