Extending Inherited Metabolic Disorder diagnostics with Biomarker Interaction Visualizations

Abstract

Background: Pyrimidine (and purine) metabolism provides essential high-energy vehicles, universal throughout all life, which carry both activated metabolites that serve as fuel and building blocks for countless metabolic processes as well as the messenger molecules that steer these processes. Furthermore, these nitrogen-containing structures encode the alphabet needed to construct the genetic code. The pyrimidine pathway overlaps with the Urea Cycle through a common metabolite (carbamoyl phosphate), with the urea cycle being responsible for the production of several amino acids and removing ammonia. Various Inherited Metabolic Disorders (IMDs) disrupt these pathways, with considerably large phenotypic variation in affected patients, hindering diagnosis. Moreover, the biomarkers for these disorders overlap substantially between the IMDs and their underlying biological pathways, interfering with identifying the deficient protein. We developed a framework that allows us to combine clinical and theoretical biomarkers with pathway models through network approaches and semantic web technologies to support the diagnosis of pyrimidine and urea cycle IMDs. Methods: We integrated literature and expert knowledge into machine-readable pathway models for pyrimidine and urea cycle disorders, including disease information and relevant downstream biomarkers. The theoretical change for each biomarker per disease was compiled based on a manual database search. The top three pathways of interest were retrieved through semantic web technologies, by selecting pathways that covered most unique markers as well as showing overlap with the theoretical marker data. These pathways and the corresponding clinical data were visualized through network analysis. Two expert laboratory scientists evaluated our approach by diagnosing a cohort of sixteen previously diagnosed patients with various pyrimidine and urea cycle disorders, based on a visualization of theoretical biomarker overlap with patient data, as well as top three pathways displaying the relationships between the individual biomarkers. Results: The number of relevant biomarkers for each patient varies greatly (five to 48), and likewise the pathways covering most unique biomarkers differ for equivalent disorders. The two experts reached similar conclusions regarding the diagnosis of nine patient samples without knowledge about clinical symptoms or sex. For the remaining seven cases, four interpretations pointed in the direction of a subset of disorders, which could be prioritized for further investigation. Three cases were found to be undiagnosable with the data available. Conclusion: The presented workflow supports the diagnosis of several IMDs of pyrimidine and urea cycle pathways, by directly linking biological pathway knowledge and theoretical biomarker data to clinical cases. This workflow is adaptable to analyze different types of IMDs, difficult patient cases and functional assays in the future. Furthermore, the pathway models can be used as a basis to perform various other types of (omics) data analysis, e.g. transcriptomics, metabolomics, fluxomics. Data Availability: The data used in this study have been deposited at https://github.com/BiGCAT-UM/IMD-PUPY .