Project Summary Kidney transplantation is the preferred treatment of patients with end-stage renal disease compared to dialysis in terms of patient survival, quality of life and cost. One of the most common causes of premature graft loss after kidney transplantation is alloimmune-mediated injury. HLA alloantigens represent a significant barrier to long-term allograft outcome. Kidney allograft failure is often caused by recipient immune recognition of foreign human leukocyte antigen (HLA) proteins of the donor organ. The HLA gene complex comprises multiple loci and has a high degree of genetic polymorphism. HLA amino acid (AA) polymorphisms strongly impact key structural and functional features of HLA molecules, including allorecognition by T cells and alloantibody. Previous studies have shown that mismatches (MMs) at HLA antigens are associated with worse outcomes, with the highest risk of graft failure (GF) particularly associated with HLA-DRB1 MMs. However, the relative impact of AA variation is unknown because organ allocation systems have not collected comprehensive molecular HLA typing data. Single-center studies using high resolution HLA typing to evaluate MMs at surface- exposed amino acids (termed “eplets”) have shown that the number of MMs correlates with the presence of de novo donor specific antibodies (dnDSA) and GF. However, these studies involved relatively few subjects and were not representative of the ethnically diverse transplant population. In addition, these studies assumed monotonic risk increase, a gap that we plan to address in this grant application. We will use HLA imputation methods that we have developed to unlock the capability to utilize the SRTR database for association analysis of AA MM categories with graft failure. In addition, we will validate and further investigate AA MM associations using a large multi-center cohort of kidney transplants wherein high resolution HLA class I and class II typing is readily available. We will also evaluate associations of HLA AA MM assortments with risk of dnDSA development. This project aims to answer several unresolved questions about HLA AA MMs and outcomes: (1) Which HLA loci are most important to match? (2) On top of antigen-level mismatches, can amino acid level mismatches further stratify outcomes? (3) Are some assortments of AA or AA motifs more important to match for than others? Our team has developed a machine learning feature engineering to discover and optimize AA MM groupings (bins) associated with GF and dnDSA. Improved patient risk stratification may help identify which transplant recipients would benefit from less aggressive immunosuppressive regimens and by reducing the number of repeat transplants due to graft failure with a poorly matched donor. Our approach generalizes broadly to other organ transplantation and possibly to hematopoietic cell transplantation.