Snowflake: A deep learning-based human leukocyte antigen matching algorithm considering allele-specific surface accessibility

HLA Mismatches Identified by a Novel Algorithm Predict Risk of Antibody-mediated Rejection From De Novo Donor-specific Antibodies

BACKGROUND

The development of de novo donor-specific antibodies (dnDSA) and antibody-mediated rejection (AMR) remains a barrier to long-term graft and patient survival. Most dnDSA are directed against mismatched donor HLA-DQ antigens. Here, we describe a novel algorithm, which we have termed categorical amino acid mismatched epitope, to evaluate HLA-DQ mismatches.

METHODS

In this algorithm, amino acid residues of HLA-DQ protein were categorized into 4 groups based on their chemical characteristics. The likelihood of categorically mismatched peptides presented by the recipient's HLA-DRB1 was expressed as a normalized value, %Rank score. Categorical HLA-DQ mismatches were analyzed in 386 heart transplant recipients who were mismatched with their donors at the HLA-DQB1 locus.

RESULTS

We found that the presence of DQB1 mismatches with %Rank score ≤1 was associated with the development of dnDSA ( P  = 0.002). Furthermore, dnDSA increased the risk of AMR only in recipients who had DQ mismatches with %Rank score ≤1 (hazard ratio = 5.8), but the freedom from AMR was comparable between recipients with dnDSA and those without dnDSA if %Rank scores of DQ mismatching were >1.

CONCLUSIONS

These results suggest that HLA-DQ mismatches evaluated by the categorical amino acid mismatched epitope algorithm can stratify the risk of development of dnDSA and AMR in heart transplant recipients.

De Novo donor-specific anti-HLA antibody risk stratification in kidney transplantation using a combination of B cell and T cell molecular mismatch assessment

Introduction

The presence of de novo donor-specific antibody (dnDSA) has detrimental effect on allograft outcomes in kidney transplantation. As humoral responses in transplantation are elicited targeting non-self-epitopes on donor HLA proteins, assessing HLA mismatches at the molecular level provides a refined means for immunological risk stratification.

Methods

In the present study, we utilized three HLA molecular mismatch assessment algorithms, Snow, HLAMatchmaker, and PIRCHE-II, to evaluate the independent and synergistic association of B cell and T cell epitope mismatches with dnDSA development in a cohort of 843 kidney transplant recipients.

Results

Our results demonstrated that B cell and T cell epitope mismatches at HLA Class I and DRB1/DQB1 loci are remarkably increased in dnDSA-positive recipients, even after normalization by allele mismatch numbers in individual study subjects. Furthermore, elevated Snow, verified eplet mismatches, and PIRCHE-II scores are significantly associated with dnDSA occurrence individually and in combination.

Conclusion

Our findings highlight the value of utilizing B cell and T cell epitope mismatch evaluation in living donor selection and immunological risk stratification to improve transplant outcomes.

Advancing risk stratification in kidney transplantation: integrating HLA-derived T-cell epitope and B-cell epitope matching algorithms for enhanced predictive accuracy of HLA compatibility

Introduction

The immune-mediated rejection of transplanted organs is a complex interplay between T cells and B cells, where the recognition of HLA-derived epitopes plays a crucial role. Several algorithms of molecular compatibility have been suggested, each focusing on a specific aspect of epitope immunogenicity.

Methods

Considering reported death-censored graft survival in the SRTR dataset, we evaluated four models of molecular compatibility: antibody-verified Eplets, Snow, PIRCHE-II and amino acid matching. We have statistically evaluated their co-dependency and synergistic effects between models systematically on 400,935 kidney transplantations using Cox proportional hazards and XGBoost models.

Results

Multivariable models of histocompatibility generally outperformed univariable predictors, with a combined model of HLA-A, -B, -DR matching, Snow and PIRCHE-II yielding highest AUC in XGBoost and lowest BIC in Cox models. Augmentation of a clinical prediction model of pre-transplant parameters by molecular compatibility metrics improved model performance particularly considering long-term outcomes.

Discussion

Our study demonstrates that the use of multiple specialized molecular HLA matching predictors improves prediction performance, thereby improving risk classification and supporting informed decision-making in kidney transplantation.

Molecular matching tools for allocation and immunosuppression optimization. Ready for primetime?

PURPOSE OF REVIEW

Molecular matching continues to be an important topic in organ transplantation. Over the years, several studies - larger and smaller - supported correlations of molecular incompatibility loads and clinical outcomes. However, their practical utility for clinical decision making remains controversial and there is no consensus on the context in which they should be used.

RECENT FINDINGS

The recent literature on molecular matching can be divided into four main areas of research: several groups present improvements of the algorithmic pipelines (1), increasing the robustness of previous findings. Further clinical evidence is reported (2) in various cohorts and other organ transplant domains, such as liver and lung transplantation. Consideration is given to the application of molecular matching in the allocation of deceased organs (3), suggesting options to improve allocation equity and utility. Furthermore, evidence is provided for personalized immunosuppression based on immunological risk (4), including infection and post graft failure management.

SUMMARY

There is ample evidence that current molecular matching algorithms add value to immunologic risk stratification for organ transplant recipients. First studies on how to translate these insights into patient management with respect to organ allocation and personalized medicine are underway and require further support.

Subthreshold rejection activity in many kidney transplants currently classified as having no rejection

Most kidney transplant patients who undergo biopsies are classified as having no rejection based on consensus thresholds. However, we hypothesized that because these patients have normal adaptive immune systems, T cell-mediated rejection (TCMR) and antibody-mediated rejection (ABMR) may exist as subthreshold activity in some transplants currently classified as no rejection. To examine this question, we studied genome-wide microarray results from 5086 kidney transplant biopsies (from 4170 patients). An updated molecular archetypal analysis designated 56% of biopsies as no rejection. Subthreshold molecular TCMR and/or ABMR activity molecular activity was detectable as elevated classifier scores in many biopsies classified as no rejection, with ABMR activity in many TCMR biopsies and TCMR activity in many ABMR biopsies. In biopsies classified as no rejection histologically and molecularly, molecular TCMR classifier scores correlated with increases in histologic TCMR features and molecular injury, lower estimated glomerular filtration rate, and higher risk of graft loss, and molecular ABMR activity correlated with increased glomerulitis and donor-specific antibody. No rejection biopsies with high subthreshold TCMR or ABMR activity had a higher probability of having TCMR or ABMR, respectively, diagnosed in a future biopsy. We conclude that many kidney transplant recipients have unrecognized subthreshold TCMR or ABMR activity, with significant implications for future problems.

HLA-EpiCheck: novel approach for HLA B-cell epitope prediction using 3D-surface patch descriptors derived from molecular dynamic simulations

Motivation

The human leukocyte antigen (HLA) system is the main cause of organ transplant loss through the recognition of HLAs present on the graft by donor-specific antibodies raised by the recipient. It is therefore of key importance to identify all potentially immunogenic B-cell epitopes on HLAs in order to refine organ allocation. Such HLAs epitopes are currently characterized by the presence of polymorphic residues called "eplets". However, many polymorphic positions in HLAs sequences are not yet experimentally confirmed as eplets associated with a HLA epitope. Moreover, structural studies of these epitopes only consider 3D static structures.

Results

We present here a machine-learning approach for predicting HLA epitopes, based on 3D-surface patches and molecular dynamics simulations. A collection of 3D-surface patches labeled as Epitope (2117) or Nonepitope (4769) according to Human Leukocyte Antigen Eplet Registry information was derived from 207 HLAs (61 solved and 146 predicted structures). Descriptors derived from static and dynamic patch properties were computed and three tree-based models were trained on a reduced non-redundant dataset. HLA-Epicheck is the prediction system formed by the three models. It leverages dynamic descriptors of 3D-surface patches for more than half of its prediction performance. Epitope predictions on unconfirmed eplets (absent from the initial dataset) are compared with experimental results and notable consistency is found.

Availability and implementation

Structural data and MD trajectories are deposited as open data under doi: 10.57745/GXZHH8. In-house scripts and machine-learning models for HLA-EpiCheck are available from https://gitlab.inria.fr/capsid.public_codes/hla-epicheck.

Quantifying uncertainty of molecular mismatch introduced by mislabeled ancestry using haplotype-based HLA genotype imputation

Introduction

Modern histocompatibility algorithms depend on the comparison and analysis of high-resolution HLA protein sequences and structures, especially when considering epitope-based algorithms, which aim to model the interactions involved in antibody or T cell binding. HLA genotype imputation can be performed in the cases where only low/intermediate-resolution HLA genotype is available or if specific loci are missing, and by providing an individuals' race/ethnicity/ancestry information, imputation results can be more accurate. This study assesses the effect of imputing high-resolution genotypes on molecular mismatch scores under a variety of ancestry assumptions.

Methods

We compared molecular matching scores from "ground-truth" high-resolution genotypes against scores from genotypes which are imputed from low-resolution genotypes. Analysis was focused on a simulated patient-donor dataset and confirmed using two real-world datasets, and deviations were aggregated based on various ancestry assumptions.

Results

We observed that using multiple imputation generally results in lower error in molecular matching scores compared to single imputation, and that using the correct ancestry assumptions can reduce error introduced during imputation.

Discussion

We conclude that for epitope analysis, imputation is a valuable and low-risk strategy, as long as care is taken regarding epitope analysis context, ancestry assumptions, and (multiple) imputation strategy.

Homozygosity in any HLA locus is a risk factor for specific antibody production: the taboo concept 2.0

Objective

In a cooperative study of the University Hospital Leipzig, University of Leipzig, and the Charité Berlin on kidney transplant patients, we analysed the occurrence of HLA-specific antibodies with respect to the HLA setup of the patients. We aimed at the definition of specific HLA antigens towards which the patients produced these antibodies.

Methods

Patients were typed for the relevant HLA determinants using mainly the next-generation technology. Antibody screening was performed by the state-of-the-art multiplex-based technology using microspheres coupled with the respective HLA alleles of HLA class I and II determinants.

Results

Patients homozygous for HLA-A*02, HLA-A*03, HLA-A*24, HLA-B*07, HLA-B*18, HLA-B*35, HLA-B*44, HLA-C*03, HLA-C*04, and HLA-C*07 in the class I group and HLA-DRB1*01, HLA-DRB1*03, HLA-DRB1*07, HLA-DRB1*15, HLA-DQA1*01, HLA-DQA1*05, HLA-DQB1*02, HLA-DQB1*03(7), HLA-DQB1*06, HLA-DPA1*01, and HLA-DPB1*04 in the class II group were found to have a significant higher antibody production compared to the heterozygous ones. In general, all HLA determinants are affected. Remarkably, HLA-A*24 homozygous patients can produce antibodies towards all HLA-A determinants, while HLA-B*18 homozygous ones make antibodies towards all HLA-B and selected HLA-A and C antigens, and are associated with an elevation of HLA-DRB1, parts of DQB1 and DPB1 alleles. Homozygosity for the HLA class II HLA-DRB1*01, and HLA-DRB1*15 seems to increase the risk for antibody responses against most of the HLA class I antigens (HLA-A, HLA-B, and HLA-C) in contrast to HLA-DQB1*03(7) where a lower risk towards few HLA-A and HLA-B alleles is found. The widely observed differential antibody response is therefore to be accounted to the patient's HLA type.

Conclusion

Homozygous patients are at risk of producing HLA-specific antibodies hampering the outcome of transplantation. Including this information on the allocation procedure might reduce antibody-mediated immune reactivity and prevent graft loss in a patient at risk, increasing the life span of the transplanted organ.

The transformative potential of artificial intelligence in solid organ transplantation

Solid organ transplantation confronts numerous challenges ranging from donor organ shortage to post-transplant complications. Here, we provide an overview of the latest attempts to address some of these challenges using artificial intelligence (AI). We delve into the application of machine learning in pretransplant evaluation, predicting transplant rejection, and post-operative patient outcomes. By providing a comprehensive overview of AI's current impact, this review aims to inform clinicians, researchers, and policy-makers about the transformative power of AI in enhancing solid organ transplantation and facilitating personalized medicine in transplant care.

PIRCHE-II Risk and Acceptable Mismatch Profile Analysis in Solid Organ Transplantation

To optimize outcomes in solid organ transplantation, the HLA genes are regularly compared and matched between the donor and recipient. However, in many cases a transplant cannot be fully matched, due to widespread variation across populations and the hyperpolymorphism of HLA alleles. Mismatches of the HLA molecules in transplanted tissue can be recognized by immune cells of the recipient, leading to immune response and possibly organ rejection. These adverse outcomes are reduced by analysis using epitope-focused models that consider the immune relevance of the mismatched HLA.PIRCHE, an acronym for Predicted Indirectly ReCognizable HLA Epitopes, aims to categorize and quantify HLA mismatches in a patient-donor pair by predicting HLA-derived T cell epitopes. Specifically, the algorithm predicts and counts the HLA-derived peptides that can be presented by the host HLA, known as indirectly-presented T cell epitopes. Looking at the immune-relevant epitopes within HLA allows a more biologically relevant understanding of immune response, and provides an expanded donor pool for a more refined matching strategy compared with allele-level matching. This PIRCHE algorithm is available for analysis of single transplantations, as well as bulk analysis for population studies and statistical analysis for comparison of probability of organ availability and risk profiles.

Repeated local ellipsoid protrusion supplements HLA surface characterization

Allorecognition of donor HLA is a major risk factor for long-term kidney graft survival. Although several molecular matching algorithms have been proposed that compare physiochemical and structural features of the donors' and recipients' HLA proteins in order to predict their compatibility, the exact underlying mechanisms are still not fully understood. We hypothesized that the ElliPro approach of single ellipsoid fitting and protrusion ranking lacks sensitivity for the characteristic shape of HLA molecules and developed a prediction pipeline named Snowball that is fitting smaller ellipsoids iteratively to substructures. Aggregated protrusion ranks of locally fitted ellipsoids were calculated for 712 publicly available HLA structures and 78 predicted structures using AlphaFold 2. Amino-acid sequence and protrusion ranks were used to train deep neural network predictors to infer protrusion ranks for all known HLA sequences. Snowball protrusion ranks appear to be more sensitive than ElliPro scores in fine parts of the HLA such as the helix structures forming the HLA binding groove in particular when the ellipsoids are fitted to substructures considering atoms within a 15 Å radius. A cloud-based web service was implemented based on amino-acid matching considering both protein- and position-specific surface area and protrusion ranks extending the previously presented Snowflake prediction pipeline.

NephroCAGE-German-Canadian Consortium on AI for Improved Kidney Transplantation Outcome: Protocol for an Algorithm Development and Validation Study

BACKGROUND

Recent advances in hardware and software enabled the use of artificial intelligence (AI) algorithms for analysis of complex data in a wide range of daily-life use cases. We aim to explore the benefits of applying AI to a specific use case in transplant nephrology: risk prediction for severe posttransplant events. For the first time, we combine multinational real-world transplant data, which require specific legal and technical protection measures.

OBJECTIVE

The German-Canadian NephroCAGE consortium aims to develop and evaluate specific processes, software tools, and methods to (1) combine transplant data of more than 8000 cases over the past decades from leading transplant centers in Germany and Canada, (2) implement specific measures to protect sensitive transplant data, and (3) use multinational data as a foundation for developing high-quality prognostic AI models.

METHODS

To protect sensitive transplant data addressing the first and second objectives, we aim to implement a decentralized NephroCAGE federated learning infrastructure upon a private blockchain. Our NephroCAGE federated learning infrastructure enables a switch of paradigms: instead of pooling sensitive data into a central database for analysis, it enables the transfer of clinical prediction models (CPMs) to clinical sites for local data analyses. Thus, sensitive transplant data reside protected in their original sites while the comparable small algorithms are exchanged instead. For our third objective, we will compare the performance of selected AI algorithms, for example, random forest and extreme gradient boosting, as foundation for CPMs to predict severe short- and long-term posttransplant risks, for example, graft failure or mortality. The CPMs will be trained on donor and recipient data from retrospective cohorts of kidney transplant patients.

RESULTS

We have received initial funding for NephroCAGE in February 2021. All clinical partners have applied for and received ethics approval as of 2022. The process of exploration of clinical transplant database for variable extraction has started at all the centers in 2022. In total, 8120 patient records have been retrieved as of August 2023. The development and validation of CPMs is ongoing as of 2023.

CONCLUSIONS

For the first time, we will (1) combine kidney transplant data from nephrology centers in Germany and Canada, (2) implement federated learning as a foundation to use such real-world transplant data as a basis for the training of CPMs in a privacy-preserving way, and (3) develop a learning software system to investigate population specifics, for example, to understand population heterogeneity, treatment specificities, and individual impact on selected posttransplant outcomes.

INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID)

DERR1-10.2196/48892.

Snowflake epitope matching correlates with child-specific antibodies during pregnancy and donor-specific antibodies after kidney transplantation

Development of donor-specific human leukocyte antigen (HLA) antibodies (DSA) remains a major risk factor for graft loss following organ transplantation, where DSA are directed towards patches on the three-dimensional structure of the respective organ donor's HLA proteins. Matching donors and recipients based on HLA epitopes appears beneficial for the avoidance of DSA. Defining surface epitopes however remains challenging and the concepts underlying their characterization are not fully understood. Based on our recently implemented computational deep learning pipeline to define HLA Class I protein-specific surface residues, we hypothesized a correlation between the number of HLA protein-specific solvent-accessible interlocus amino acid mismatches (arbitrarily called Snowflake) and the incidence of DSA. To validate our hypothesis, we considered two cohorts simultaneously. The kidney transplant cohort (KTC) considers 305 kidney-transplanted patients without DSA prior to transplantation. During the follow-up, HLA antibody screening was performed regularly to identify DSA. The pregnancy cohort (PC) considers 231 women without major sensitization events prior to pregnancy who gave live birth. Post-delivery serum was screened for HLA antibodies directed against the child's inherited paternal haplotype (CSA). Based on the involved individuals' HLA typings, the numbers of interlocus-mismatched antibody-verified eplets (AbvEPS), the T cell epitope PIRCHE-II model and Snowflake were calculated locus-specific (HLA-A, -B and -C), normalized and pooled. In both cohorts, Snowflake numbers were significantly elevated in recipients/mothers that developed DSA/CSA. Univariable regression revealed significant positive correlation between DSA/CSA and AbvEPS, PIRCHE-II and Snowflake. Snowflake numbers showed stronger correlation with numbers of AbvEPS compared to Snowflake numbers with PIRCHE-II. Our data shows correlation between Snowflake scores and the incidence of DSA after allo-immunization. Given both AbvEPS and Snowflake are B cell epitope models, their stronger correlation compared to PIRCHE-II and Snowflake appears plausible. Our data confirms that exploring solvent accessibility is a valuable approach for refining B cell epitope definitions.