Predicting Parkinson’s: Using Neural Networks to Evaluate the Genetic Risk Factors Behind the Disease

Abstract

To date, researchers do not know the exact reasons for the loss of dopaminergic neurons in the substantia nigra pars compacta that leads to Parkinson’s Disease (PD). Thus, it is extremely difficult to predict whether or not a patient is likely to develop the disease later on, as their risk increases with age. However, once patients present with the common symptoms indicative of the illness, a substantial amount of dopaminergic neurons are already lost. Seeing as there are no current avenues of replacing those neurons, predictive diagnosis and preventive measures could be of extraordinary help in devising treatments. Our aim was to use the significant research into possible high-risk genetic factors from genome-wide association studies (GWAS) to formulate a predictive neural network model for Parkinson’s. We analyzed patient genomes for mutations in the top 20 genes associated with PD, as well as 21 genes implicated in axon guidance pathways, to determine whether the patients were at high or low risk for Parkinson’s. Our model produced an accuracy and AUROC of 94%. We found this significant because it showed a strong correlation between the single nucleotide polymorphisms (SNPs) we analyzed and PD. We believe our model can be further improved upon by adding considerations for other investigated risk factors, such as patient age, familial history of disease, or gut microbiota inconsistencies among others.