Results
The improvement in concordance between the empirical and reported quality scores of SNVs is shown in Fig. 2A. A significant improvement in the accuracy of quality scores after recalibration is shown in Fig. 2B. The association test for NK-AML yielded 42 SNVs that had a p-value of less than 0.05 and passed the quality threshold of PLINK/SEQ (data not shown). Among 42 SNVs, we excluded 11 SNVs with no rs ID, four SNVs with call rates less than 1, and five synonymous SNVs for subsequent analysis.
A total of 21 nsSNVs located in 18 candidate genes were selected and included in the GRS models after considering pairwise linkage disequilibrium (Table 1). This study replicated three somatic mutations of the MUC4, CNTNAP2, and GNAS genes, which were reported in previous studies on leukemia [9, 10, 18, 19]. In addition, we identified novel point mutations in 15 other genes. The SNVs rs75156964 (ATP13A3; odds ratio [OR], 9.33; p = 0.013) and rs56213454 (GNAS; OR, 9.33; p = 0.013) showed the most significant evidence for association, and rs6604516 (PPIAL4G; OR, 21.00; p = 0.016) showed the strongest effect size among the 21 SNVs. Other SNVs also showed strong effects in the risk of NK-AML (OR, 9.33 to 13.5).
We compared the prediction accuracy of 21 stepwise GRS models. The predictive power for NK-AML reached 100% in model 5, which consisted of five nsSNVs: rs75156964 (ATP13A3, 3q29), rs56213454 (GNAS, 20q13.3), rs6604516 (PPIAL4G, 1q21.1), rs10888338 (OR2T33, 1q44), and rs2443878 (ANKRD36, 2q11.2) (Table 2, Fig. 3A). The five SNVs (rs75459784, rs55824312, rs200239604, rs56213454, and rs56371919) in three previously reported genes-MUC4, CNTNAP2, and GNAS-also showed a high AUC value (AUC, 0.98; 95% confidence interval [CI], 0.92 to 1.00).