A deep learning method for HLA imputation and trans-ethnic MHC fine-mapping of type 1 diabetes

Naito, Tatsuhiko; Suzuki, Ken; Hirata, Jun; Kamatani, Yoichiro; Matsuda, Koichi; Toda, Tatsushi; Okada, Yukinori

doi:10.1038/s41467-021-21975-x

Cited by 61 publications

(63 citation statements)

References 63 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Given the growing diversity of HLA alleles, deep learning could help increase the number of alleles that could be imputed accurately when more data is available [34]. Recently, Naito et al independently published an effective multitask CNN architecture for HLA imputation that is very similar to the one proposed in this work [26]. This work confirms the effectiveness of this deep learning approach, and provides an additional occlusion analysis.…”

Section: Introductionsupporting

confidence: 67%

HLA Allele Imputation with Multitask Deep Convolutional Neural Network

Chen¹

2021

Preprint

View full text Add to dashboard Cite

The Human leukocyte antigen (HLA) system is a highly polymorphic gene complex encoding the major histocompatibility complex proteins in humans. HLA alleles are of strong epidemiological interest for their large effect sizes in associations with autoimmune diseases, infectious diseases, severe drug reactions, and transplant medicine. Since HLA genotyping can be time-consuming and cost-prohibitive, methods to impute HLA alleles from SNP genotype data have been developed, including HLA Genotype Imputation with Attribute Bagging (HIBAG), HLA*IMP:02, and SNP2HLA. However, limitations of these imputation programs include imputation accuracy, computational runtime, and ability to impute HLA allele haplotypes. We present a deep learning framework for HLA allele imputation using a multitask convolutional neural network (CNN) architecture. In this approach, we use phased SNP genotype data flanking ±250 kb from each HLA locus to simultaneously impute HLA allele haplotyes across loci HLA-A, -B, -C, -DQA1, -DQB1, -DPA1, -DPB1, and -DRB1. We start by tokenizing phased genotype sequences into k-mers that serve as input to the model. The CNN architecture starts with a shared embedding layer for learning low-dimensional representations of k-mers, shared convolutional layers for detecting genotype motifs, and branches off into separate densely-connected layers for imputing each HLA loci. We present evidence that the CNN used information from known tag SNPs to impute HLA alleles, and demonstrate the architecture is robust against a selection of hyperparameters. On the T1DGC dataset, our model achieved 97.6% imputation accuracy, which was superior to SNP2HLA's performance and comparable to HIBAG's performance. However, unlike HIBAG, our method can impute an entire HLA haplotype sequence instead of imputing one locus at a time. Additionally, by separating the training and inference steps, our imputation program provides user flexibility to reduce usage time.

show abstract

Section: Introductionsupporting

confidence: 67%

HLA Allele Imputation with Multitask Deep Convolutional Neural Network

Chen¹

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Given their critical impact on immune responses and contribution to host genetics of various infectious diseases, 26,27 HLA gene variants have been investigated for their possible role in the response to COVID-19 infection with controversial discussions 28,29 . To address this issue, we applied in silico imputation of both classical and non-classical HLA variants using the HLA reference panel of Japanese ( n = 1,118) 30,31 . After imputing the HLA variants, we did not observe association signals satisfying neither of the genome-wide significance ( P < 5.0 × 10 -8 ) or HLA-wide significance thresholds ( P < 0.05/2,482 variants = 2.0 × 10 -5 ; Supplementary Figure 3 and Supplementary Table 7 ).…”

Section: Resultsmentioning

confidence: 99%

“…HLA genotype imputation was performed using DEEP*HLA software (version 1.0), a multitask convolutional deep learning method 31 . We used the population-specific imputation reference panel of Japanese ( n = 1,118), which included both classical and non-classical HLA gene variants for imputation 30 .…”

Section: Methodsmentioning

confidence: 99%

Japan COVID-19 Task Force: a nation-wide consortium to elucidate host genetics of COVID-19 pandemic in Japan

Namkoong

Edahiro

Fukunaga

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

To elucidate the host genetic loci affecting severity of SARS-CoV-2 infection, or Coronavirus disease 2019 (COVID-19), is an emerging issue in the face of the current devastating pandemic. Here, we report a genome-wide association study (GWAS) of COVID-19 in a Japanese population led by the Japan COVID-19 Task Force, as one of the initial discovery GWAS studies performed on a non-European population. Enrolling a total of 2,393 cases and 3,289 controls, we not only replicated previously reported COVID-19 risk variants (e.g., LZTFL1, FOXP4, ABO, and IFNAR2), but also found a variant on 5p35 (rs60200309-A at DOCK2) that was significantly associated with severe COVID-19 in younger (<65 years of age) patients with a genome-wide significant p-value of 1.2 × 10-8 (odds ratio = 2.01, 95% confidence interval = 1.58-2.55). This risk allele was prevalent in East Asians, including Japanese (minor allele frequency [MAF] = 0.097), but rarely found in Europeans. Cross-population Mendelian randomization analysis made a causal inference of a number of complex human traits on COVID-19. In particular, obesity had a significant impact on severe COVID-19. The presence of the population-specific risk allele underscores the need of non-European studies of COVID-19 host genetics.

show abstract

“… 20 Furthermore, MHC fine‐mapping for a trans‐ethnic cohort can boost its power of revealing genetic features that affect complex diseases beyond ethnicities by removing confounding by linkage. 21 , 22 We performed trans‐ethnic MHC fine‐mapping of PD using GWAS data in European and East Asian populations and identified that specific amino acid positions of HLA‐DRβ1 and HLA‐B were independently associated with PD risk across ethnicities. Furthermore, the risk‐associated alleles of HLA‐DRB1 presented variable binding affinity to a known α‐synuclein epitope in in silico prediction, suggesting their functional role to the pathogenesis of PD.…”

mentioning

confidence: 99%

Trans‐Ethnic Fine‐Mapping of the Major Histocompatibility Complex Region Linked to Parkinson's Disease

et al. 2021

Self Cite

View full text Add to dashboard Cite

A BS TRACT: Background: Despite evidence for the role of human leukocyte antigen (HLA) in the genetic predisposition to Parkinson's disease (PD), the complex haplotype structure and highly polymorphic feature of the major histocompatibility complex (MHC) region has hampered a unified insight on the genetic risk of PD. In addition, a majority of the reports focused on Europeans, lacking evidence on other populations. Objectives: The aim of this study is to elucidate the genetic features of the MHC region associated with PD risk in trans-ethnic cohorts. Methods: We conducted trans-ethnic fine-mapping of the MHC region for European populations (14,650 cases and 1,288,625 controls) and East Asian populations (7712 cases and 27,372 controls). We adopted a hybrid fine-mapping approach including both HLA genotype imputation of genome-wide association study (GWAS) data and direct imputation of HLA variant risk from the GWAS summary statistics.Results: Through trans-ethnic MHC fine-mapping, we identified the strongest associations at amino acid position 13 of HLA-DRβ1 (P = 6.0 × 10 −15 ), which explains the majority of the risk in HLA-DRB1. In silico prediction revealed that HLA-DRB1 alleles with histidine at amino acid position 13 (His13) had significantly weaker binding affinity to an α-synuclein epitope than other alleles (P = 9.6 × 10 −4 ). Stepwise conditional analysis suggested additional independent associations at Ala69 in HLA-B (P = 1.0 × 10 −7 ). A subanalysis in Europeans suggested additional independent associations at non-HLA genes in the class III MHC region (EHMT2; P = 2.5 × 10 −7 ). Conclusions: Our study highlights the shared and distinct genetic features of the MHC region in patients with PD across ethnicities.

show abstract

A deep learning method for HLA imputation and trans-ethnic MHC fine-mapping of type 1 diabetes

Cited by 61 publications

References 63 publications

HLA Allele Imputation with Multitask Deep Convolutional Neural Network

HLA Allele Imputation with Multitask Deep Convolutional Neural Network

Japan COVID-19 Task Force: a nation-wide consortium to elucidate host genetics of COVID-19 pandemic in Japan

Trans‐Ethnic Fine‐Mapping of the Major Histocompatibility Complex Region Linked to Parkinson's Disease

Contact Info

Product

Resources

About