Table 1.
Accuracy and call rate for a 2/3 (training data) – 1/3 (validation data) cross-validation experiment, using a call threshold of T = 0.7 .
Locus | Number of validated | Call rate (2-digit) | Accuracy (2-digit) | Call rate (4-digit) | Accuracy (4-digit) |
---|---|---|---|---|---|
HLA-A | 816 | 0.98 | 0.98 | 0.98 | 0.97 |
HLA-B | 1009 | 0.98 | 0.98 | 0.98 | 0.96 |
HLA-C | 635 | 0.98 | 0.97 | 0.97 | 0.97 |
HLA-DRB1 | 858 | 0.99 | 0.98 | 0.95 | 0.92 |
HLA-DQA1 | 51 | 1 | 0.98 | 0.98 | 0.98 |
HLA-DQB1 | 867 | 1 | 0.99 | 0.99 | 0.98 |
User datasets are always imputed using the full set of training data, which should result in greater accuracy.