Automated Verification of Phenotypes using PubMed

Ryan Bridges, Jette Henderson, Joyce C. Ho, Byron C. Wallace, Joydeep Ghosh
2016 Proceedings of the 7th ACM International Conference on Bioinformatics, Computational Biology, and Health Informatics - BCB '16  
In the realm of data driven clinical research, medical concepts, or phenotypes, are used to serve as indicators for patient clusters of interest. Often, studies will use groups of algorithmically generated phenotypes (feature groups) to predict the occurrence of heart disease, diabetes, and other conditions. When these groups are algorithmically generated, the most common method of verification is manual human annotation, which can be time consuming and sometimes inconsistent. In this paper, we
more » ... propose a supervision-free method of verification that uses co-occurrence in PubMed articles to determine clinical significance. We demonstrate the efficacy of the method by 1) applying it to phenotypes generated through automatic machine learning methods and 2) showing it separates randomly generated groups of phenotypes from curated groups of known, clinical phenotypes.
doi:10.1145/2975167.2985844 dblp:conf/bcb/BridgesHHWG16 fatcat:err3ixhcszbhhfsz5oad4cnrhy