Applying Familial Hypercholesterolemia diagnostic approach to diseases where a Polygenic Risk Score have been reported
The original scope was the validation and improvement of our Machine Learning approach to analyze interactions in your genotyping/imputation data for better Familial Hypercholesterolemia diagnostics. It turned out for us that the prediction models based on our feature selection tools show better accuracies than the established Polygenic Risk Score (PRS). PRS have limitations as they take only additive effects into account. Here, we are convinced to have something better than PRS.
Now, we are faced with the question whether our method only works for Familial Hypercholesterolemia or for all kind diseases and traits. Therefore, we wish to apply our analyzing approach to other diseases where a Polygenic Risk Score have been reported. Our aim is to benchmark our approach against as many existing Polygenic Risk Scores in order to provide a better prediction tool for the future.
Therefore, we select Polygenic Risk Scores for diseases with reported prediction values such as area accuracy, sensitivity, or are under curve (AUC) calculated with UK Biobank data. That enables us to re-run the analysis of UK Biobank data with our approach and, finally, compare the prediction values of our results with reported PRS from the literature. From there, we select diseases with reported PRS built with UK Biobank data by other users. We hope to develop better prediction models based on our interaction-based approach than the existing additive PRS to bring real benefit for disease prediction & prevention.