Each ontology was scored for perceived value (PV) by the 9 members of the project
team, from numerous pharmaceutical and biotechnology companies. Each ontology
was assigned a score of 3 for high PV, 2 for medium PV and 1 for low PV and 0 for no
PV by each of the 9 team members. This gave the total PV score (a simple summation
of scores) for each of the 54 mappings predicted by Paxo, which informed our priorities
for evaluation:
The parameters of Paxo were selected to balance recall (matches missing from the
LOOM baseline standard) and precision (correct matches from random sampling from
unique matches where n=60). Recall ranged from 66% to 97% while precision for
unique matches ranged from 45% to 95% for each mapping. These predicted mapping
sets will be made accessible openly via the project web page [
Fifty-four ontology mappings were predicted using the Paxo algorithm which demonstrates how it can be applied to any pair of ontologies hosted by OLS and OxO at EMBL-EBI, within a single domain where overlap of class concepts is likely to be found.
As no hand-curated gold standard mappings exist to measure recall, in the near future
we will use a panel of numerous algorithms to generate a set of silver standard
mappings from a minimum of three consensus votes as we have published previously
[
Future work may include crowd validation of predicted mappings and further mapping between ontologies in the clinical domain.
We would like to express gratitude to all the Pistoia Alliance Ontology Mapping project team members and their parent organisations who contributed expertise, time and funding. EJR was supported by the AIDA project, funded by the Alan Turing Institute, and the SIRIUS Centre for Scalable Data Access (Research Council of Norway, project no.: 237889).