Computational Approaches for Clinical, Genomic and Proteomic Markers of Response to GLP-1 Therapy in Type-2 Diabetes Mellitus: An Exploratory Analysis With Machine Learning Algorithms.

Purpose: The International Diabetes Federation reported that 463 million adults had diabetes in 2019, a number expected to reach 700 million by 2045. Glucagon-like peptide-1 agonists (GLP-1) offer various benefits in managing diabetes but about 40% of patients don’t respond well to this therapy, leading to sustained high HbA1c levels and complications. This study aims to understand the response to GLP-1 therapy among individuals achieving HbA1c levels ≤ 53mmol/mol, using clinical, genomic, proteomic data and create a Machine Learning (ML) based predictor for GLP-1 responsiveness.

Methods: Here, we report analysis from a T2DM dataset of individuals prescribed with GLP-1 therapy from the Diastrat cohort recruited at the Northern Ireland Centre for Stratified Medicine (NICSM), in Northern Ireland.

Results: The novelty of this study is that we have identified statistically significant (p-value <=0.05) genomic variants rs6724083 (NCKAP5), rs7596517 (NCKAP5), rs7593167 (NCKAP5), rs12911054 (DPH6-AS1) and rs2569431 (CTU1) which appear to differentiate responders and non-responders and have identified 45 statistically significant (p-value <=0.05, q-value <=0.05) proteomic markers that can differentiate healthy controls, GLP-1 responders, GLP-1 non-responders and GLP-1 treatment groups out of which ITGA11, HAOX1, ANG1, PAI and PDGF had fold change ~2. A proteomic machine learning model was built; the best model predicted response with 95% classification accuracy. These results change the direction of type-2 diabetes research in biomarker and knowledge discovery useful for research and can inform clinical practice.

Conclusion: Further testing in prospective validation cohorts is required to determine the clinical utility of the proposed model.