Disease-based Phenotypical Gene Priorization

Wolpers, Anja (2024) Disease-based Phenotypical Gene Priorization. Masters thesis, Institute for Visual and Analytic Computing, University of Rostock.

Text Master_Anja_Wolpers_Disease-based _Phenotypical_Gene_Priorization.pdf - Published Version Restricted to Registered users only Download (3MB)

Abstract

Medical experts investigate potentially disease-causing base pairs in a patient's DNA called genetic variants to diagnose genetic diseases. As every human DNA has thousands of variants - most of which are harmless - it is too time-consuming to look for the variant causing the patient's symptoms by hand. Therefore, algorithms for phenotypical gene prioritization compare the patient's symptoms with symptoms that can be caused by variants in each gene based on data from the human phenotype ontology (HPO). Thus, for all genes, the semantic similarity between the patient's and the gene's symptoms is calculated and the genes can be sorted - prioritized - by their symptom's similarity. This way, the variants in genes that are more likely to cause the patient's symptoms and, therefore, the patient's disease are examined first. This makes the diagnosis of genetically caused diseases much more efficient in most cases. The varvis® software, a clinical decision support system developed by Limbus Medical Technologies™ GmbH (Limbus) includes an algorithm for phenotypical gene prioritization that is routinely used by many of its users and helps them in their everyday diagnostics. In this thesis, I first adapted Limus' algorithm to compare the symptoms based on diseases instead of genes because most genes cause more than one disease, but patients generally only have one single disease. Then, I developed another approach that takes into account that most patients only have a subset of the symptoms associated with a disease. Therefore, it weights the symptoms' similarities with the symptoms' frequencies in patients with a disease. I analyzed the two approaches using anonymous real-life data provided by Limbus and discussed how they could be implemented and improved further in practice.