BOE: BiOnt Enhanced (DrugProt corpus)

The motivation behind this project was to get better results, on the DrugProt corpus (drug and chemical-protein interactions), through the adaptation of the BiOnt system.

Installation

Pursue the steps of requirements.txt in the folders: bin; data; drugprot-evaluation-library.

Enhancements

• Ontology mapping and Ancestor linking are multiprocessed;
• Added faster sentence segmenter (pysbd);
• GO and ChEBI entities ancestors are populated through the .db file and via API;
• Creation of additional senteces with the overlapped entities;
• Improved text tokenization;
• Optimization of some functions;
• Fixed minor bugs.

Results (Micro-averaged)

Principal variations between the models shown below are as follows:

baseline: Model generated using a modified version of the BiOnt system used in the BioCreative VII Track 1 challenge.

main ents: The largest term from the previously identified overlapping entities is used, while the rest are discarded.

extra ents: Creates new sentences that incorporate one GENE term and one CHEMICAL term mentioned in the original sentence.

Model	Precision	Recall	F1-Score
baseline	0.618	0.097	0.168
main ents	0.24	0.293	0.264
extra ents	0.149	0.262	0.19
main ents (updated)	0.444	0.447	0.445
extra ents (updated)	0.449	0.431	0.44

Micro-averaged results of the three models predicting the relations of the DrugProt development dataset.

Preprocessing

• $2: type_of_action (preprocess)
• $3: preprocessing_method (extra or main)
• $4: temporary_directory

Example:

python src/ontologies_embeddings.py preprocess main temp/

python src/ontologies_embeddings.py preprocess main temp_dev/

Training

• $2: type_of_action (train)
• $3: model_name
• $4: channels (words; wordnet; concatenation_ancestors)
• $5: temporary_directory

Example:

python src/ontologies_embeddings.py train model_name words wordnet concatenation_ancestors temp/

Predicting

• $2: type_of_action (test)
• $3: model_name
• $4: channels (words; wordnet; concatenation_ancestors)
• $5: temporary_directory

Example:

python src/ontologies_embeddings.py test model_name words wordnet concatenation_ancestors temp_dev/

Evaluating

Coverting the result file into a friendly format to be evaluated

• $2: path to Prediction TXT file
• $3: path to new Prediction TSV file

Example:

python scripts/validation.py results/model_name_results.txt results/model_name_results.tsv

Using the DrugProt Evaluation library

• $2: -g/--gs_path: path to Gold Standard relations TSV file
• $3: -p/--pred_path: path to Prediction TSV file
• $4: -e/--ent_path: path to Gold Standard entities TSV file
• $5: --pmids: path to list of relevant PMIDs

Example:

python drugprot-evaluation-library/src/main.py -g gs-data/drugprot_development_relations.tsv -p results/model_name_results.tsv -e gs-data/drugprot_development_entities.tsv --pmids gs-data/pmids.txt