Back

Machine Learning Insights into HLA Noncoding Sequence Mismatches and Their Impact on DPB1 Matching in Hematopoietic Cell Transplantation

Askar, M.; Mosbruger, T. L.; Shaw, G. T.-W.; Kim, H.; Duan, Y.; Allen, A. S.; Duke, J. L.; Olson, T. S.; Monos, D. S.; Hayeck, T. J.

2024-09-14 hematology
10.1101/2024.09.13.24313580 medRxiv
Show abstract

PurposeHCT is vital for treating hematological malignancies, relying on HLA matching between unrelated patient-donor pairs to significantly reduce adverse outcomes. Recent studies recognize the potential impact of HLA-DPB1 mismatches on HCT outcomes. Multiple approaches focus on finding better-tolerated HLA-DPB1 mismatches. Additionally, recent studies suggest matching at noncoding HLA sequence may improve HCT outcomes. This study aims to evaluate different approaches for categorizing DPB1 mismatches in patient-donor pairs and explore the potential impact of noncoding mismatches (available in class I HLA genes) on clinical outcomes. MethodsA retrospective study of 5,106 HCT pairs using Cox proportional hazards models, weighted by a machine learning algorithm, evaluates the impact of particular combinations of HLA-DPB1 mismatches in the context of noncoding HLA class I mismatches on outcomes of HCT. HLA-DPB1 mismatch criteria included T-cell epitope permissive/non-permissive mismatches, expression markers, and evolutionary clade mismatches. ResultsTwo HLA-DPB1 mismatches, using multiple criteria, lead to significant hazards of acute graft versus host disease grades 2-4, in the T cell replete group. When HLA-DPB1 mismatches occurred across evolutionary clades (DP2 allele/low-expression patient vs DP5 allele/high-expression in the donor), the deplete group showed significant hazards for treatment-related mortality (TRM) (HR=1.94, p-value=8.9x10-7) and overall survival (OS) (HR=1.67, p-value=1.3x10-5) for additive effects of noncoding mismatches with two HLA-DPB1 mismatches. ConclusionTwo HLA-DPB1 mismatches remain to predict worse outcomes. However, noncoding mismatches in HLA class I genes confer elevated hazards of TRM and OS in conjunction with mismatches across evolutionary branches of HLA-DPB1. Therefore, noncoding mismatches may inform donor selection in the presence of HLA-DPB1 mismatches and improve HCT outcomes, emphasizing the utility of comprehensive sequencing of HLA alleles in HCT settings.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.