Improved Type 2 Diabetes Risk Stratification in the Qatar Biobank Cohort by Ensemble Learning Classifier Incorporating Multi-Trait, Population-Specific, Polygenic Risk Scores

Abstract

Background: Type 2 Diabetes (T2D) is a pervasive chronic disease influenced by a complex interplay of environmental and genetic factors. To enhance T2D risk prediction, leveraging genetic information is essential, with polygenic risk scores (PRS) offering a promising tool for assessing individual genetic risk. Our study focuses on the comparison between multi-trait and single-trait PRS models and demonstrates how the incorporation of multi-trait PRS into risk prediction models can significantly augment T2D risk assessment accuracy and effectiveness. Methods: We conducted genome-wide association studies (GWAS) on 12 distinct T2D-related traits within a cohort of 14,278 individuals, all sequenced under the Qatar Genome Programme (QGP). This in-depth genetic analysis yielded several novel genetic variants associated with T2D, which served as the foundation for constructing multiple weighted PRS models. To assess the cumulative risk from these predictors, we applied machine learning (ML) techniques, which allowed for a thorough risk assessment. Results: Our research identified genetic variations tied to T2D risk and facilitated the construction of ML models integrating PRS predictors for an exhaustive risk evaluation. The top-performing ML model demonstrated a robust performance with an accuracy of 0.8549, AUC of 0.92, AUC-PR of 0.8522, and an F1 score of 0.757, reflecting its strong capacity to differentiate cases from controls. We are currently working on acquiring independent T2D cohorts to validate the efficacy of our final model. Conclusion: Our research underscores the potential of PRS models in identifying individuals within the population who are at elevated risk of developing T2D and its associated complications. The use of multi-trait PRS and ML models for risk prediction could inform early interventions, potentially identifying T2D patients who stand to benefit most based on their individual genetic risk profile. This combined approach signifies a stride forward in the field of precision medicine, potentially enhancing T2D risk prediction, prevention, and management.