Harnessing the Human Phenotype Ontology to Predict the Age of Onset of Rare Genetic Diseases

Abstract

In this research, we employ machine learning techniques to predict the age of onset of rare genetic diseases using the Human Phenotype Ontology. Provid- ing age of onset information to rare genetic diseases may assist clinicians in differential diagnosis of patients, by narrowing down results. We first employed a random forest regression model, followed up by a graph convolutional network in an effort to capture temporal traits and nuanced re- lationships within the dataset. While both models performed well above the baseline with a top-1 accuracy of 85 and 84%, respectively, we failed to identify a significant increase in performance of the neural network over the random forest model. With this research, we highlight the potential of machine learning in differ- ential diagnostics by capturing the relationship between phenotypic traits and disease progression.