gms | German Medical Science

Introduction: In precision oncology and biomedical cancer research the support of tools is needed to select the best drugs based on the patients’ or cell lines’ genetic background. The challenge is to normalize the different annotations, i.e., substance, drug name or drug class to the same level, most often to the drug class. A manual curated classification is time consuming and may be biased towards the indented effect. Existing resources do not cover all levels, are missing granularity or unify drug class and target into the same classification. Decision support and large data research will benefit from a drug ontology that is built automatically and covers the characteristics of a drug in a structured way.

Methods: Here we introduce DrugOn, a drug ontology that incorporates information from multiple public data sources. By accessing and processing data from the Drugbank database [1] and other drug classification systems such as ATC codes and MeSH terms, we create a new ontology combining the most relevant drug-related information based on these sources. Using a selection process, we can identify the best options in terms of drug classes and categories, molecular targets as well as outlinks and other annotations. In this way, the most appropriate drug categories and targets can be identified for each drug based on the data sources.

Results: 192 drugs found in two lymphoma research projects served as test cases for the evaluation and were annotated by an expert. These drugs are assigned to 117 classes and 380 targets in the automated classification. Compared to manual classification, DrugOn had equivalent drug classes for 154 out of 192 oncology drugs. In 20 cases, no output was possible, due to missing data from public sources. DrugOn can be queried via a REST API. In addition, a front-end application provides the ability to view and discover the ontology and make individual queries about drugs.

Discussion: DrugOn is an automatically built drug ontology based on public data sources completed by some processing rules. The result is an unbiased and easy to update resource for oncological drugs. The ontology distinguishes between drug class and the target protein. Through the structured ontology format and the API DrugOn can be included in manual or automated pipelines.

Conclusion: DrugOn has its origin in lymphoma projects with large amount of data. It can be easily extended to other cancer entities. Possible extensions are to include more sources, e.g., to also include descriptions of gene functions. In the current form DrugOn is already used in research projects and in the variant interpretation framework Onkopus for molecular tumor boards. The front-end version is available at https://mtb.bioinf.med.uni-goettingen.de/drugon-web.

Kevin Kornrumpf and Vera Gnass: equal contribution

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.