Classification of genetic developmental disorders in the
                         peer-reviewed literature using biomedical ontologies -
                         Abstract
                         T. Michael Yates1,∗ , T. Ian Simpson1,2
                         1
                           Institute for Adaptive and Neural Computation, Informatics Forum, The University of Edinburgh, 10 Crichton Street, Edinburgh
                         EH8 9AB, UK
                         2
                           Simons Initiative for the Developing Brain, The University of Edinburgh, Hugh Robson Building, George Square, Edinburgh EH8
                         9XF, UK


                                      Abstract
                                      Diagnosis in genetic developmental disorders (GDD) requires analysis of phenotypic information from thousands
                                      of conditions described in the peer-reviewed literature. Automated search and classification of these manuscripts is
                                      needed to enable phenotype data extraction at scale. Here, we describe a classifier developed for this purpose, using
                                      features derived from several biomedical ontologies. PubMed search strategies were evaluated to identify filters
                                      which enriched for GDD-relevant manuscripts. This was used to inform the creation of a corpus of 13,000 papers
                                      describing 230 exemplar GDD. The title and abstract were manually annotated to papers containing phenotypic
                                      data for GDD or not. A random forest binary classification algorithm was trained on this dataset, using features
                                      derived from manuscript metadata, MeSH (Medical Subject Headings) ontology, Human Phenotype Ontology
                                      and Mondo Disease Ontology. The PubMed gene symbol[title] filter was selected as the best balance between
                                      minimizing large numbers of results (>100,000 for some genes) and enriching for GDD-relevant manuscripts.
                                      36 features were used, with the top 5 by feature importance being: disease name in abstract, case report in
                                      publication type, disease name in title, published in GDD journal, and ‘mutation’ in MeSH terms. The precision
                                      was 0.81 and recall 0.83. We demonstrate effective classification of the GDD peer-reviewed literature using
                                      supervised machine learning. Application of this system, in combination with named entity recognition, will
                                      allow for creation of phenotypic data models across the spectrum of GDD.

                                      Keywords
                                      Ontology, genetic developmental disorders, machine learning, phenotype analysis




                          15th International Conference on Biomedical Ontologies 2024, July 17-19, 2024, Enschede, The Netherlands
                         ∗
                              Corresponding author.
                          Envelope-Open tyates@ed.ac.uk (T. M. Yates); ian.simpson@ed.ac.uk (T. I. Simpson)
                          Orcid 0000-0002-0118-5946 (T. M. Yates); 0000-0003-0495-7187 (T. I. Simpson)
                                     © 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).


CEUR
                  ceur-ws.org
Workshop      ISSN 1613-0073
Proceedings