Abstract
Donor/recipient mismatched HLA antigens can lead to the production of Donor-Specific Antibodies by the recipient, which are deleterious to organ transplants. The HLA-DQ locus is the most frequent target, with both the DQ beta and alpha chains involved. For deceased donors in particular, while HLA-DQB1 has been typed in emergencies for a long time, HLA-DQA1 has only recently been included. No imputation algorithmic tool was available to impute HLA-DQA1 until the development of HaploSFHI, trained on 61,393 two-field typings by NGS methods. We evaluated the accuracy of two-field HLA-DQA1 imputation from serological and two-field level HLA-A, B, DRB1, and DQB1 typings. We report a highly accurate two-field HLA-DQA1 prediction using a French test cohort of 7696 individuals, respectively reaching 92.30% and 96.45% accuracy. The average 'False Positive eplet load' stood at 0.19 and 0.07, respectively, and the average 'False Negative eplet load' at 0.18 and 0.08, respectively. A similar performance was obtained on three independent test cohorts of European ancestry (from the USA, the UK, and Portugal). Interestingly, performance was only slightly inferior on five independent test cohorts of other ethnicities (from Hong Kong and the USA) whereas it was significantly lower for two-field DRB1 imputation from its serological level. These results suggest that DQA1 can reliably be imputed even when information is totally missing, with low error risk at both antigen and eplet levels, even if the reference population is not matched. Similar additional initiatives would be welcome to confirm these findings.