Definition of High-risk Type 1 Diabetes HLA-DR and HLA-DQ Types Using Only Three Single Nucleotide Polymorphisms

Evaluating risk of developing type 1 diabetes (T1D) depends on determining an individual's HLA type, especially of the HLA DRB1 and DQB1 alleles. Individuals positive for HLA-DRB1*03 (DR3) or HLA-DRB1*04 (DR4) with DQB1*03:02 (DQ8) have the highest risk of developing T1D. Currently, HLA typing methods are relatively expensive and time consuming. We sought to determine the minimum number of single nucleotide polymorphisms (SNPs) that could rapidly define the HLA-DR types relevant to T1D, namely, DR3/4, DR3/3, DR4/4, DR3/X, DR4/X, and DRX/X (where X is neither DR3 nor DR4), and could distinguish the highest-risk DR4 type (DR4-DQ8) as well as the non-T1D-associated DR4-DQB1*03:01 type. We analyzed 19,035 SNPs of 10,579 subjects (7,405 from a discovery set and 3,174 from a validation set) from the Type 1 Diabetes Genetics Consortium and developed a novel machine learning method to select as few as three SNPs that could define the HLA-DR and HLA-DQ types accurately. The overall accuracy was 99.3%, area under curve was 0.997, true-positive rates were >0.99, and false-positive rates were <0.001. We confirmed the reliability of these SNPs by 10-fold cross-validation. Our approach predicts HLA-DR/DQ types relevant to T1D more accurately than existing methods and is rapid and cost-effective.