Download This Paper
Preprints with The Lancet is a collaboration between The Lancet Group of journals and SSRN to facilitate the open sharing of preprints for early engagement, community comment, and collaboration. Preprints available here are not Lancet publications or necessarily under review with a Lancet journal. These preprints are early-stage research papers that have not been peer-reviewed. The usual SSRN checks and a Lancet-specific check for appropriateness and transparency have been applied. The findings should not be used for clinical or public health decision-making or presented without highlighting these facts. For more information, please see the FAQs.
Articlespolygenic Risk Score for Predicting the Development of Autoimmune Diseases Using Graph Convolutional Networks: A Model Development and Validation Study
27 Pages Posted: 18 Jan 2024
More...Abstract
BackgroundExisting linear polygenic risk score (PRS) models for autoimmune diseases have limitations in accurately accounting for non-additive risk effects and interactions between HLA haplotypes, key aspects in these diseases' genetic composition. Graph convolutional networks (GCN) can address these challenges by integrating both the genotypes of risk variants and genotypic relationships among samples. However, there's a lack of extensive research validating the diagnostic capabilities of GCN-based PRS across multiple diseases. MethodsWe utilized the genotypic data of four autoimmune diseases (rheumatoid arthritis (RA), multiple sclerosis (MS), psoriasis (PSO), celiac disease (CEL)) and two non-autoimmune diseases (Atrial Fibrillation, Alzheimer's diesase) from the UK Biobank. We constructed models on a cohort of British Whites using 5-fold cross-validation and tested their performance on non-British Whites. For RA, we also used the BioBank Japan and IORRA. The performance of the GCN models were compared to those constructed by PRSice-2, PRScs, and a Simple neural net using AUC and Nagelkerke’s R2. FindingsWhen constructed using SNPs from the HLA region, the GCN models consistently outperformed other PRS models in all comparisons for each of the four autoimmune diseases with p-values <0.05. However, GCN was comparable to other models for non-autoimmune diseases. Additionally, the GCN model demonstrated 20-50% higher Nagelkerke’s R2 values compared to other methods for all four autoimmune diseases. Visualization of the graph generated by the GCN revealed connected networks formed based on the count of HLA risk alleles. InterpretationThe results underscore GCN' role in significantly improving PRS accuracy for autoimmune diseases, particularly those linked to the HLA region. This advancement suggests a substantial leap in precision medicine, emphasizing GCN' ability to decode complex genetic patterns. FundingJSPS KAKENHI, AMED, the Joint Usage/Research Program of Medical Research Institute Tokyo Medical and Dental University, and the Medical Research Center Initiative for High Depth Omics.
Keywords: polygenic risk scores, radiographic progression, autoimmune disease, prediction, deep learning
Suggested Citation: Suggested Citation