Using predicate and provenance information from a knowledge graph for drug efficacy screening.

Area of Science:

• Biomedical informatics
• Computational biology
• Machine learning in drug discovery

Background:

• Biomedical knowledge graphs (BKGs) model complex biological information using subject-predicate-object triples.
• BKGs are valuable for drug efficacy screening but often overlook predicate and provenance data.
• Predicate and provenance information offer insights into relationships and data sources within BKGs.

Purpose of the Study:

• To develop a supervised machine learning classifier for drug efficacy screening.
• To evaluate the added value of predicate and provenance information in BKGs.
• To ensure biological relevance by focusing on protein-level drug-target and disease-protein interactions.

Main Methods:

• Utilized random forests with repeated 10-fold cross-validation for classification.
• Developed a novel approach incorporating predicate and provenance information from BKGs.
• Benchmarked against a state-of-the-art method that excludes predicate and provenance data.

Main Results:

• The proposed method achieved an Area Under the ROC Curve (AUC) of 78.1% and 74.3% on two reference datasets.
• Outperformed a state-of-the-art knowledge graph technique by achieving AUCs of 65.6% and 64.6%.
• Classifiers using both predicate and provenance information yielded the best performance, surpassing those using either alone.

Conclusions:

• Predicate and provenance information significantly enhance the accuracy of drug efficacy screening.
• The developed machine learning classifier demonstrates the added value of these data types.
• This approach offers a more robust method for identifying potential drug-disease relationships.