Measuring ambiguity in HLA typing methods

Comparison Between Antigen and Allelic HLA Mismatches, and the Risk of Acute Rejection in Kidney Transplant Recipients

Deceased donor kidney allocation relies on HLA compatibility at the antigen level, as optimal matching reduces the risk of acute rejection. Whether HLA allele-level mismatches improve, the prediction of acute rejection after transplantation remains unclear. Using data from the Australia and New Zealand Dialysis and Transplant Registry (ANZDATA) from 2017 to 2020, HLA antigenic and allelic mismatches between recipients and deceased donors were calculated with imputation of two-field allelic equivalents undertaken where required. The discordance between antigen and allele mismatches was calculated, and oblique random survival forest models were used to predict acute rejection. Predictive performance of antigen (HLA-A, -B, -DRB1 and -DQB1), allele (HLA-A, -B, -DRB1 and -DQB1) and extended allele (HLA-A, -B, -C, -DRB1, -DQA1 and -DQB1) models was examined using concordance index and integrated Brier scores, with variable importance calculated using permutation-based methods. Among 2644 recipients followed for a median of 1.7 years, 521 recipients (20%) experienced acute rejection. Discordant numbers of antigenic and allelic mismatches occurred in 8%, 9%, 24% and 17% of HLA-A, -B, -DRB1 and -DQB1 loci, respectively. Predictive performances were similar across all models, with concordance indices of 0.62-0.63 and integrated Brier scores of 0.09. HLA-DRB1 and -DQB1 mismatches were the strongest predictors of acute rejection across models. In patients matched at the HLA-DRB1 or -DQB1 antigen, those with allelic mismatches had similar incidences of rejection compared to those without. Allelic-level assessment of HLA compatibility did not improve the prediction of acute rejection and may disadvantage certain recipients by reclassifying them into higher mismatch categories in allocation algorithms without providing clear clinical benefit.

Can We Use Eplets (or Molecular) Mismatch Load Analysis to Improve Organ Allocation? The Hope and the Hype

In recent years, there have been calls for implementation of "epitope matching" in deceased-donor organ allocation policies (later changed to "eplet matching"). Emerging data indeed support the use of molecular mismatch load analysis in specific patient groups, with the objective of posttransplant stratification into different treatment arms. For this purpose, the expectation is to statistically categorize patients as low- or high-immune-risk. Importantly, these patients will continue to be monitored' and their risk category, as well as their management, can be adjusted according to on-going findings. However, when discussing deceased donor organ allocation and matching algorithms, where the decision is not modifiable and has lasting impact on outcomes, the situation is fundamentally different. The goal of changing allocation schemes is to achieve the best possible HLA compatibility between donor and recipient. Immunologically speaking, this is a very different objective. For this purpose, the specific interplay of immunogenicity between the donor and any potential recipient must be understood. In seeking compatibility, the aim is not to redefine matching but to identify those mismatches that are "permissible" or' in other words, less immunogenic. In our eagerness to improve transplant outcome, unfortunately, we have conflated the hype with the hope. Terminology is used improperly, and new terms are created in the process with no sufficient support. Here, we call for a cautious evaluation of baseline assumptions and a critical review of the evidence to minimize unintended consequences.

Predictive Models for Recurrent Membranous Nephropathy After Kidney Transplantation

Recurrent membranous nephropathy (MN) posttransplantation affects 35% to 50% of kidney transplant recipients (KTRs) and accounts for 50% allograft loss 5 y after diagnosis. Predictive factors for recurrent MN may include HLA-D risk alleles, but other factors have not been explored with certainty.

The Accuracy of Sequence-Specific Oligonucleotide and Real-Time Polymerase Chain Reaction HLA Typing in Determining the Presence of Pre-Transplant Donor-Specific Anti-HLA Antibodies and Total Eplet Mismatches for Deceased Donor Kidney Transplantation

High resolution human leukocyte antigen (HLA) typing is important in establishing eplet compatibility and the specificity of donor-specific anti-HLA antibodies (DSA). In deceased donor kidney transplantation, high resolution donor HLA typing may not be immediately available, leading to inaccuracies during the organ allocation process. We aimed to determine the concordance and agreement of HLA-Class I and II eplet mismatches calculated using population frequency based allelic haplotype association (linkage disequilibrium, LD) from sequence-specific oligonucleotide (SSO) and real-time polymerase chain reaction (rtPCR) donor HLA typing (available at time of donor kidney allocation) compared to high-resolution Next Generation Sequencing (NGS) donor typing. NGS high resolution HLA typing were available for all recipients prior to donor kidney allocation. A cohort of 94 deceased donor-recipient pairs from a single Western Australian center were included (77 individual donors typed, 55 local and 22 interstate). The number of class I (HLA-A+B+C) and class II (HLA-DRB1+DRB3/4/5+DQB1+DQA1+DPB1+DPA1) eplet mismatches were calculated using HLAMatchmaker, comparing LD- and NGS-HLA typing. The accuracy in assigning pre-transplant DSA was compared between methods. The concordance correlation coefficient (95%CI) for HLA-class I and II eplet mismatches were 0.994 (0.992 to 0.996) and 0.991 (0.986 to 0.993), respectively. The 95% limits of agreement for class I were -1.3 (-1.6 to -1.1) to 1.4 (1.2 to 1.7) and -4.8 (-5.7 to -3.9) to 5.0 (4.1 to 5.9) for Class II. Disagreement between the two methods were present for 11 and 37 of the Class I and II donor/recipient pairs. Of which, 5 had a difference of ≥5 class II eplet mismatches. There were 34 (36%) recipients with potential pre-transplant DSA, of which 8 (24% of recipients with DSA) had indeterminate and ultimately false positive DSA assigned by donor LD-typing. While the concordance between NGS- and LD-typing was high, the limits of agreement suggest meaningful differences between these two techniques. The inaccurate assignment of DSA from donor LD-typing may result in associated HLA being considered unacceptable mismatches, inappropriately precluding candidates’ access to transplantation. Accurate imputation of two-field HLA alleles based on LD from SSO and rtPCR HLA typing remains a substantial challenge in clinical practice in-lieu of widely available, rapid, high-resolution methods.

Impact of HLA Epitope Matching on Outcomes After Unrelated Bone Marrow Transplantation

The significance of antibody-identified epitopes stimulating humoral alloimmunity is not well understood in the identification of non-permissive human leukocyte antigen (HLA) mismatching patterns in hematopoietic stem cell transplantation (HSCT). This was a retrospective study in a cohort of 9,991 patients who underwent their first HSCT for hematologic malignancies from unrelated bone marrow donors in the Transplant Registry Unified Management Program (TRUMP). HLA eplet mismatches (EMM) were quantified using HLAMatchmaker (HLAMM). The median age of patients was 48 years (range, 16 to 77). The number of EMM in recipient-donor pairs in our study population ranged from 0 to 37 in HLA class I (median, 0) and 0 to 60 in HLA class II (median, 1). In addition to the known high-risk mismatch patterns in the Japanese cohort, HLA-C EMM in the GVH direction was associated with a significantly higher risk for grade III-IV aGVHD, leading to a higher risk of non-relapse mortality and lower overall survival (compared with HLA-C matched patients, HR 1.67, 95% CI 1.44–1.95; HR 1.39, 95% CI 1.25–1.54; HR 1.20, 95% CI 1.10–1.30, respectively). HLAMM-based epitope matching might be useful for identifying patients who are at high risk for serious complications after HSCT from HLA mismatched unrelated donors.

Improving Access to HLA-Matched Kidney Transplants for African American Patients

Introduction

Kidney transplants fail more often in Black than in non-Black (White, non-Black Hispanic, and Asian) recipients. We used the estimated physicochemical immunogenicity for polymorphic amino acids of donor/recipient HLAs to select weakly immunogenic kidney transplants for Black vs. White or non-Black patients.

Methods

OPTN data for 65,040 donor/recipient pairs over a 20-year period were used to calculate the individual physicochemical immunogenicity by hydrophobic, electrostatic and amino acid mismatch scores (HMS, EMS, AMS) and graft-survival outcomes for Black vs. White or vs. non-Black recipients, using Kaplan-Meier survival and Cox regression analyses. Simulations for re-matching recipients with donors were based on race-adjusted HMS thresholds with clinically achievable allocations.

Results

The retrospective median kidney graft survival was 12.0 years in Black vs. 18.6 years in White (6.6-year difference; p>0.001) and 18.4 years in non-Black (6.4-year difference; p>0.01) recipients. Only 0.7% of Blacks received transplants matched at HLA-A/B/DR/DQ (HMS=0) vs. 8.1% in Whites (p<0.001). Among fully matched Blacks (HMS=0), graft survival was 16.1-years and in well-matched Blacks (HMS ≤ 3.0) it was 14.0-years. Whites had 21.6-years survival at HMS ≤ 3.0 and 18.7-years at HMS ≤ 7.0 whereas non-Blacks had 22.0-year at HMS ≤ 3.0 and 18.7-year at HMS ≤ 7.0, confirming that higher HMS thresholds produced excellent survival. Simulation of ABO-compatible donor-recipient pairs using race-adjusted HMS thresholds identified weakly immunogenic matches at HMS=0 for 6.1% Blacks and 18.0% at HMS ≤ 3.0. Despite prioritizing Black patients, non-Black patients could be matched at the same level as in current allocation (47.0% vs 56.5%, at HMS ≤ 7.0).

Conclusions

Race-adjusted HMS (EMS, AMS)-based allocation increased the number of weakly immunogenic donors for Black patients, while still providing excellent options for non-Black recipients.

hlaR: A rapid and reproducible tool to identify eplet mismatches between transplant donors and recipients

Eplet mismatch load, both overall and at the single molecule level, correlates with transplant recipient outcomes. However, precise eplet assessment requires high-resolution HLA typing of both the donor and recipient. Anything less than high-resolution typing requires imputation of HLA types. The currently available methods to identify eplet mismatch are both tedious and demanding. Therefore, we developed a software package and user-friendly web application (hlaR), that simplifies the workflow of eplet analysis, provides functions to impute high-resolution from low-resolution data and calculates both overall and single molecule eplet mismatch for single or multiple donor recipient pairs. Compared to manual assessments using currently available tools (namely, HLAMatchMaker), hlaR resulted in only minimal discrepancy in eplet mismatches (mean absolute difference of 0.56 for class I and 0.86 for class II for unique sum across loci). Additionally, output of the single molecule eplet function compared well to manual calculation, with an average single antigen count increase of 0.19. Importantly, the hlaR tool permits rapid and reproducible imputation and eplet mismatch including comparison between eplet reference tables (e.g. HLAMatchMaker version 2 or 3). Users can import data from a spreadsheet rather than relying on keystroke entry of individual donor and recipient data, thus reducing the risk of data entry errors. The resulting improved scalability of the hlaR tool is highlighted by plotting analysis time against the size of the input dataset. The new hlaR tool can provide eplet mismatch data with a streamlined workflow. With decreased effort from the end user, eplet matching and mismatch load data can be further incorporated into both research and clinical use.

HLA-A, -B, -C, -DRB1 allele and haplotype frequencies of the Korean population and performance characteristics of HLA typing by next-generation sequencing

INTRODUCTION

Human leukocyte antigen (HLA) identification at the allelic level is important for haematopoietic stem cell transplantation (HSCT). Next-generation sequencing (NGS) resolves ambiguous alleles by determining the phase of the polymorphisms. The aim of this study was to validate the software for HLA-SBT (sequence-based typing), assess Korean allele frequency, and characterise the performance of NGS-HLA typing.

METHODS

From the 2009 to 2016 registry, 1293 unrelated healthy donors with a complete dataset of previously characterised HLA-A, -B, -C, and -DRB1 loci were selected and assessed for frequency, haplotype inference, and relative linkage disequilibrium. For performance characteristics of NGS-HLA, alleles included in 1293 cases and ambiguous or alleles assigned as new by SBT-HLA software, or unassigned alleles were included. A total of 91 and 41 quality control samples resulted in 1056 alleles (132 samples × 4 loci × 2 diploid) for analysis. The GenDx NGSgo kit was used for NGS-HLA typing using the Illumina MiSeq platform.

RESULTS

A panel of 132 samples covered 231 alleles, including 53 HLA-A, 80 HLA-B, 43 HLA-C, and 55 HLA-DRB1 by HLA-SBT typing. Comparison of SBT-HLA and NGS-HLA typing showed 99.7% (1053/1056) concordance and discrepant cases were resolved by manual evaluation. Typing by NGS resulted in 67 HLA-A, 112 HLA-B, 71 HLA-C, and 72 HLA-DRB1 alleles. A total of 132 ambiguous, 4 new, and 1 unassigned alleles by HLA-SBT were resolved by NGS-HLA typing.

CONCLUSIONS

NGS-HLA typing provided robust and conclusive results without ambiguities, and its implementation could support HSCT in clinical settings.

A blueprint for electronic utilization of ambiguous molecular HLA typing data in organ allocation systems and virtual crossmatch

Virtual crossmatch (VXM) compares a transplant candidate's unacceptable antigens to the HLA typing of the donor before an organ offer is accepted and, in selected cases, supplant a prospective physical crossmatch. However, deceased donor typing can be ambiguous, leading to uncertainty in compatibility prediction. We have developed a prototype web application that utilizes ambiguous HLA molecular typing data to predict which unacceptable antigens are present in the donor HLA genotype as donor-specific antibodies (DSA). The application compares a candidate's listed unacceptable antigens to computed probabilities of all possible two-field donor HLA alleles and UNOS antigens. The VIrtual CrossmaTch for mOleculaR HLA typing (VICTOR) tool can be accessed at http://www.transplanttoolbox.org/victor. We reanalyzed historical VXM cases where a transplant center's manual interpretation of molecular typing results influenced offer evaluation. We found that interpretation of ambiguous donor molecular typing data using imputation could one day influence VXM decisions if the DSA predictions were rigorously validated. Standardized interpretation of molecular typing data, if applied to the match run, could also change which offers are made. HLA typing ambiguity has been an underappreciated source of immunological risk in organ transplantation. The VICTOR tool can serve as a testbed for development of allocation policies with the aim of decreasing offers refused due to HLA incompatibility.

The association between HLA and non-Hodgkin lymphoma subtypes, among a transplant-indicated population

Several studies have implicated HLA in non-Hodgkin lymphoma (NHL) subtype etiology. However, NHL patients indicated for stem cell transplants are underrepresented in these reports. We therefore evaluated the association between HLA and NHL subtypes among a transplant-indicated population. One thousand three hundred and sixty-six NHL patients HLA-typed and indicated for transplant at the City of Hope National Medical Center (Duarte, CA) were compared to 10,271 prospective donors. Odds ratios and 95% confidence intervals were calculated for HLA haplotype and alleles, adjusted for sex and age. The HLA-A*0201∼C*0602∼B*1302∼DRB1*0701∼DQB1*0201 haplotype was significantly associated with follicular lymphoma (FL) risk among Caucasians. Several haplotypes were associated with diffuse large B-cell lymphoma (DLBCL) risk among Caucasians, including the previously implicated DLBCL risk loci, HLA-B*0801. The HLA-A*0101 allele was also observed to be associated with mantle cell lymphoma (MCL) risk. Our results support the association between previously reported susceptibility loci and FL and suggest potentially new DLBCL and MCL risk loci.

Evaluation of computational programs to predict HLA genotypes from genomic sequencing data

Motivation

Despite being essential for numerous clinical and research applications, high-resolution human leukocyte antigen (HLA) typing remains challenging and laboratory tests are also time-consuming and labour intensive. With next-generation sequencing data becoming widely accessible, on-demand in silico HLA typing offers an economical and efficient alternative.

Results

In this study we evaluate the HLA typing accuracy and efficiency of five computational HLA typing methods by comparing their predictions against a curated set of > 1000 published polymerase chain reaction-derived HLA genotypes on three different data sets (whole genome sequencing, whole exome sequencing and transcriptomic sequencing data). The highest accuracy at clinically relevant resolution (four digits) we observe is 81% on RNAseq data by PHLAT and 99% accuracy by OptiType when limited to Class I genes only. We also observed variability between the tools for resource consumption, with runtime ranging from an average of 5 h (HLAminer) to 7 min (seq2HLA) and memory from 12.8 GB (HLA-VBSeq) to 0.46 GB (HLAminer) per sample.

While a minimal coverage is required, other factors also determine prediction accuracy and the results between tools do not correlate well. Therefore, by combining tools, there is the potential to develop a highly accurate ensemble method that is able to deliver fast, economical HLA typing from existing sequencing data.

Different DRB1*03:01-DQB1*02:01 haplotypes confer different risk for celiac disease

Celiac disease is associated with the HLA-DR3-DQA1*05:01-DQB1*02:01 and DR4-DQA1*03:01-DQB1*03:02 haplotypes. In addition, there are currently over 40 non-HLA loci associated with celiac disease. This study extends previous analyses on different HLA haplotypes in celiac disease using next generation targeted sequencing. Included were 143 patients with celiac disease and 135 non-celiac disease controls investigated at median 9.8 years (1.4-18.3 years). PCR-based amplification of HLA and sequencing with Illumina MiSeq technology were used for extended sequencing of the HLA class II haplotypes HLA-DRB1, DRB3, DRB4, DRB5, DQA1 and DQB1, respectively. Odds ratios were computed marginally for every allele and haplotype as the ratio of allelic frequency in patients and controls as ratio of exposure rates (RR), when comparing a null reference with equal exposure rates in cases and controls. Among the extended HLA haplotypes, the strongest risk haplotype for celiac disease was shown for DRB3*01:01:02 in linkage with DQA1*05:01-DQB1*02:01 (RR = 6.34; P-value < .0001). In a subpopulation analysis, DRB3*01:01:02-DQA1*05:01-DQB1*02:01 remained the most significant in patients with Scandinavian ethnicity (RR = 4.63; P < .0001) whereas DRB1*07:01:01-DRB4*01:03:01-DQA1*02:01-DQB1*02:02:01 presented the highest risk of celiac disease among non-Scandinavians (RR = 7.94; P = .011). The data also revealed 2 distinct celiac disease risk DR3-DQA1*05:01-DQB*02:01 haplotypes distinguished by either the DRB3*01:01:02 or DRB3*02:02:01 alleles, indicating that different DRB1*03:01-DQB1*02:01 haplotypes confer different risk for celiac disease. The associated risk of celiac disease for DR3-DRB3*01:01:02-DQA1*05:01-DQB1*02:01 is predominant among patients of Scandinavian ethnicity.

Charting improvements in US registry HLA typing ambiguity using a typing resolution score

Unrelated stem cell registries have been collecting HLA typing of volunteer bone marrow donors for over 25years. Donor selection for hematopoietic stem cell transplantation is based primarily on matching the alleles of donors and patients at five polymorphic HLA loci. As HLA typing technologies have continually advanced since the beginnings of stem cell transplantation, registries have accrued typings of varied HLA typing ambiguity. We present a new typing resolution score (TRS), based on the likelihood of self-match, that allows the systematic comparison of HLA typings across different methods, data sets and populations. We apply the TRS to chart improvement in HLA typing within the Be The Match Registry of the United States from the initiation of DNA-based HLA typing to the current state of high-resolution typing using next-generation sequencing technologies. In addition, we present a publicly available online tool for evaluation of any given HLA typing. This TRS objectively evaluates HLA typing methods and can help define standards for acceptable recruitment HLA typing.

Fine-mapping of HLA associations with chronic lymphocytic leukemia in US populations

Chronic lymphocytic leukemia (CLL) displays remarkable ethnic predisposition for whites, with relative sparing of African-American and Asian populations. In addition, CLL displays among the highest familial predispositions of all hematologic malignancies, yet the genetic basis for these differences is not clearly defined. The highly polymorphic HLA genes of the major histocompatibility complex play a central role in immune surveillance and confer risk for autoimmune and infectious diseases and several different cancers, the role for which in the development of CLL has not been extensively investigated. The National Marrow Donor Program/Be The Match has collected HLA typing from CLL patients in need of allogeneic hematopoietic stem cell transplant and has recruited millions of volunteers to potentially donate hematopoietic stem cells. HLA genotypes for 3491 US white, 397 African-American, and 90 Hispanic CLL patients were compared with 50 000 controls per population from the donor registry. We identified several HLA alleles associated with CLL susceptibility in each population, reconfirming predisposing roles of HLA-A*02:01 and HLA-DRB4*01:01 in whites. Associations for haplotype DRB4*01:01∼DRB1*07:01∼DQB1*03:03 were replicated across all 3 populations. These findings provide a comprehensive assessment of the role of HLA in the development of severe CLL.

A catalog of HLA type, HLA expression, and neo-epitope candidates in human cancer cell lines

Cancer cell lines are a tremendous resource for cancer biology and therapy development. These multipurpose tools are commonly used to examine the genetic origin of cancers, to identify potential novel tumor targets, such as tumor antigens for vaccine devel-opment, and utilized to screen potential therapies in preclinical studies. Mutations, gene expression, and drug sensitivity have been determined for many cell lines using next-generation sequencing (NGS). However, the human leukocyte antigen (HLA) type and HLA expression of tumor cell lines, characterizations necessary for the development of cancer vaccines, have remained largely incomplete and, such information, when available, has been distributed in many publications. Here, we determine the 4-digit HLA type and HLA expression of 167 cancer and 10 non-cancer cell lines from publically available RNA-Seq data. We use standard NGS RNA-Seq short reads from "whole transcriptome" sequencing, map reads to known HLA types, and statistically determine HLA type, heterozygosity, and expression. First, we present previously unreported HLA Class I and II genotypes. Second, we determine HLA expression levels in each cancer cell line, providing insights into HLA downregulation and loss in cancer. Third, using these results, we provide a fundamental cell line "barcode" to track samples and prevent sample annotation swaps and contamination. Fourth, we integrate the cancer cell-line specific HLA types and HLA expression with available cell-line specific mutation information and existing HLA binding prediction algorithms to make a catalog of predicted antigenic mutations in each cell line. The compilation of our results are a fundamental resource for all researchers selecting specific cancer cell lines based on the HLA type and HLA expression, as well as for the development of immunotherapeutic tools for novel cancer treatment modalities.

A one-step DNA sequencing strategy to HLA type hematopoietic stem cell donors at recruitment - rethinking typing strategies

In order to reduce the time required to identify a match for unrelated donor hematopoietic stem cell transplantation, a one-step DNA sequencing strategy was employed at the time of recruitment. The impact of this strategy on human leukocyte antigen (HLA) typing resolution and the effect of current registry requirements on resolution and coding of assignments were evaluated. Sanger-based DNA sequencing was used to obtain diploid exons 2 and 3 HLA-A, -B and -C assignments of 2747 unrelated African American and 1822 European American volunteers at recruitment. The results demonstrate the high resolution of the approach and challenge several aspects of the current registry typing strategy. Of the 46% of African American and 74% of European American individuals whose HLA typing resulted in alternative genotypes, the majority (≥93%) was predicted to have only a single 'common' genotype among the alternatives. The common practice of adding secondary assays to resolve alternative genotype assignments that include more than two antigen groups was also evaluated. While the percentage of assignments with greater than two antigen groups reached as high as 21% (HLA-A in European Americans), only 1.8% of individuals at most carried two common genotypes encompassing three antigen groups. The assignment of (National Marrow Donor Program) NMDP-designated allele codes to the one-pass results reduced the resolution substantially and introduced genotypes that were not included in the laboratory's assignments. We suggest the alternative strategy of using the exons 2-3 diploid nucleotide sequence as the assignment submitted to the registry with the added benefit of immortalizing the assignment in time regardless of the introduction of novel alleles. To keep pace with current donor selection criteria and with the increasing number of new alleles, it is time to rethink our recruitment typing strategies.

HLA typing from RNA-seq data using hierarchical read weighting [corrected]

Correctly matching the HLA haplotypes of donor and recipient is essential to the success of allogenic hematopoietic stem cell transplantation. Current HLA typing methods rely on targeted testing of recognized antigens or sequences. Despite advances in Next Generation Sequencing, general high throughput transcriptome sequencing is currently underutilized for HLA haplotyping due to the central difficulty in aligning sequences within this highly variable region. Here we present the method, HLAforest, that can accurately predict HLA haplotype by hierarchically weighting reads and using an iterative, greedy, top down pruning technique. HLAforest correctly predicts >99% of allele group level (2 digit) haplotypes and 93% of peptide-level (4 digit) haplotypes of the most diverse HLA genes in simulations with read lengths and error rates modeling currently available sequencing technology. The method is very robust to sequencing error and can predict 99% of allele-group level haplotypes with substitution rates as high as 8.8%. When applied to data generated from a trio of cell lines, HLAforest corroborated PCR-based HLA haplotyping methods and accurately predicted 16/18 (89%) major class I genes for a daughter-father-mother trio at the peptide level. Major class II genes were predicted with 100% concordance between the daughter-father-mother trio. In fifty HapMap samples with paired end reads just 37 nucleotides long, HLAforest predicted 96.5% of allele group level HLA haplotypes correctly and 83% of peptide level haplotypes correctly. In sixteen RNAseq samples with limited coverage across HLA genes, HLAforest predicted 97.7% of allele group level haplotypes and 85% of peptide level haplotypes correctly.