MagicalRsq: Machine-learning-based genotype imputation quality calibration

Aggregating single nucleotide polymorphisms improves filtering for false-positive associations postimputation

Imputation causes bias in P-values in downstream genome-wide association studies. Imputation quality measures such as IMPUTE info are used to discriminate between false and true associations. However, implementing a high threshold often discards true associations, while a low threshold preserves false associations. This poses a challenge, especially for studies genotyped with SNP arrays. In practice, association signals register as spikes of low P-values for SNPs in close proximity owing to linkage disequilibrium, but postimputation filtering is conducted on SNPs independently. We simulated 1536 small case-control studies on the human chromosome 19 both to quantify the introduced bias and to evaluate postimputation filtering. The established IMPUTE info thresholds 0.3 and 0.8 were compared on individual SNPs and aggregated spikes in the formats "best guess genotype" and "dosage." Furthermore, we applied 2 recently published methods, Iam hiQ and MagicalRsq, to assess their effect on filtering. We found differences in false signals and imputation quality between the genotype formats, especially in the midrange between thresholds. In this midrange, 51 and 60% of associated SNPs for best guess and dosage format, respectively, are true associations. For aggregated SNPs, the majority of spikes in the midrange are true associations. We propose a new method, the Midrange Filter, which uses both thresholds and formats to classify spikes instead of SNPs. This method discards up to the same number of false signals as the upper threshold, while preserving all true associations in most simulation settings. The PsyCourse study is included as a real-data application.

Polygenic Scores of Cardiometabolic Risk Factors in American Indian Adults

Importance

Numerous efforts have been made to include diverse populations in genetic studies, but American Indian populations are still severely underrepresented. Polygenic scores derived from genetic data have been proposed in clinical care, but how polygenic scores perform in American Indian individuals and whether they can predict disease risk in this population remains unknown.

Objective

To study the performance of polygenic scores for cardiometabolic risk factors of lipid traits and C-reactive protein in American Indian adults and to determine whether such scores are helpful in clinical prediction for cardiometabolic diseases.

Design, Setting, and Participants

The Strong Heart Study (SHS) is a large American Indian cohort recruited from 1989 to 1991, with ongoing follow-up (phase VII). In this genetic association study, data from SHS American Indian participants were used in addition to data from 2 large-scale, external, ancestry-mismatched genome-wide association studies (GWASs; 450 865 individuals from a European GWAS and 33 096 individuals from a multi-ancestry GWAS) and 1 small-scale internal ancestry-matched American Indian GWAS (2000 individuals). Analyses were conducted from February 2023 to August 2024.

Exposure

Genetic risk score for cardiometabolic disease risk factors from 6 traits including 5 lipids (apolipoprotein A, apolipoprotein B, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and triglycerides), and an inflammatory biomarker (C-reactive protein [CRP]).

Main Outcomes and Measures

Data from SHS participants and the 2 GWASs were used to construct 8 polygenic scores. The association of polygenic scores with cardiometabolic disease was assessed using 2-sided z tests and 1-sided likelihood ratio tests.

Results

In the 3157 SHS participants (mean [SD] age, 56.44 [8.12] years; 1845 female [58.4%]), a large European-based polygenic score had the most robust performance (mean [SD] R2 = 5.0% [1.7%]), but adding a small-scale ancestry-matched GWAS using American Indian data helped improve polygenic score prediction for 5 of 6 traits (all but CRP; mean [SD] R2, 7.6% [3.2%]). Lipid polygenic scores developed in American Indian individuals improved prediction of diabetes compared with baseline clinical risk factors (area under the curve for absolute improvement, 0.86%; 95% CI, 0.78%-0.93%; likelihood ratio test P = 3.8 × 10-3).

Conclusions and Relevance

In this genetic association study of lipids and CRP among American Indian individuals, polygenic scores of lipid traits were found to improve prediction of diabetes when added to clinical risk factors, although the magnitude of improvement was small. The transferability of polygenic scores derived from other populations is still a concern, with implications for the advancement of precision medicine and the potential of perpetuating health disparities, particularly in this underrepresented population.

Empirical versus estimated accuracy of imputation: optimising filtering thresholds for sequence imputation

BACKGROUND

Genotype imputation is a cost-effective method for obtaining sequence genotypes for downstream analyses such as genome-wide association studies (GWAS). However, low imputation accuracy can increase the risk of false positives, so it is important to pre-filter data or at least assess the potential limitations due to imputation accuracy. In this study, we benchmarked three different imputation programs (Beagle 5.2, Minimac4 and IMPUTE5) and compared the empirical accuracy of imputation with the software estimated accuracy of imputation (Rsqsoft). We also tested the accuracy of imputation in cattle for autosomal and X chromosomes, SNP and INDEL, when imputing from either low-density or high-density genotypes.

RESULTS

The accuracy of imputing sequence variants from real high-density genotypes was higher than from low-density genotypes. In our software benchmark, all programs performed well with only minor differences in accuracy. While there was a close relationship between empirical imputation accuracy and the imputation Rsqsoft, this differed considerably for Minimac4 compared to Beagle 5.2 and IMPUTE5. We found that the Rsqsoft threshold for removing poorly imputed variants must be customised according to the software and this should be accounted for when merging data from multiple studies, such as in meta-GWAS studies. We also found that imposing an Rsqsoft filter has a positive impact on genomic regions with poor imputation accuracy due to large segmental duplications that are susceptible to error-prone alignment. Overall, our results showed that on average the imputation accuracy for INDEL was approximately 6% lower than SNP for all software programs. Importantly, the imputation accuracy for the non-PAR (non-Pseudo-Autosomal Region) of the X chromosome was comparable to autosomal imputation accuracy, while for the PAR it was substantially lower, particularly when starting from low-density genotypes.

CONCLUSIONS

This study provides an empirically derived approach to apply customised software-specific Rsqsoft thresholds for downstream analyses of imputed variants, such as needed for a meta-GWAS. The very poor empirical imputation accuracy for variants on the PAR when starting from low density genotypes demonstrates that this region should be imputed starting from a higher density of real genotypes.

A genome-wide association study of alloimmunization in the TOPMed OMG-SCD cohort identifies a locus on chromosome 12

BACKGROUND

Red cell alloimmunization after exposure to donor red cells is a very common complication of transfusion for patients with sickle cell disease (SCD), resulting frequently in accelerated donor red blood cell destruction. Patients show substantial differences in their predisposition to alloimmunization, and genetic variability is one proposed component. Although several genetic association studies have been conducted for alloimmunization, the results have been inconsistent, and the genetic determinants of alloimmunization remain largely unknown.

STUDY DESIGN AND METHODS

We performed a genome-wide association study (GWAS) in 236 African American (AA) SCD patients from the Outcome Modifying Genes in Sickle Cell Disease (OMG-SCD) cohort, which is part of Trans-Omics for Precision Medicine (TOPMed), with whole-genome sequencing data available. We also performed sensitivity analyses adjusting for different sets of covariates and applied different sample grouping strategies based on the number of alloantibodies patients developed.

RESULTS

We identified one genome-wide significant locus on chr12 (p = 3.1e-9) with no evidence of genomic inflation (lambda = 1.003). Further leveraging QTL evidence from GTEx whole blood and/or Jackson Heart Study PBMC RNA-Seq data, we identified a number of potential genes, such as ARHGAP9, STAT6, and ATP23, that may be driving the association signal. We also discovered some suggestive loci using different analysis strategies.

DISCUSSION

We call for the community to collect additional alloantibody information within SCD cohorts to further the understanding of the genetic basis of alloimmunization in order to improve transfusion outcomes.

Imputation accuracy across global human populations

Genotype imputation is now fundamental for genome-wide association studies but lacks fairness due to the underrepresentation of references from non-European ancestries. The state-of-the-art imputation reference panel released by the Trans-Omics for Precision Medicine (TOPMed) initiative improved the imputation of admixed African-ancestry and Hispanic/Latino samples, but imputation for populations primarily residing outside of North America may still fall short in performance due to persisting underrepresentation. To illustrate this point, we imputed the genotypes of over 43,000 individuals across 123 populations around the world and identified numerous populations where imputation accuracy paled in comparison to that of European-ancestry populations. For instance, the mean imputation r-squared (Rsq) for variants with minor allele frequencies between 1% and 5% in Saudi Arabians (n = 1,061), Vietnamese (n = 1,264), Thai (n = 2,435), and Papua New Guineans (n = 776) were 0.79, 0.78, 0.76, and 0.62, respectively, compared to 0.90-0.93 for comparable European populations matched in sample size and SNP array content. Outside of Africa and Latin America, Rsq appeared to decrease as genetic distances to European-ancestry reference increased, as predicted. Using sequencing data as ground truth, we also showed that Rsq may over-estimate imputation accuracy for non-European populations more than European populations, suggesting further disparity in accuracy between populations. Using 1,496 sequenced individuals from Taiwan Biobank as a second reference panel to TOPMed, we also assessed a strategy to improve imputation for non-European populations with meta-imputation, but this design did not improve accuracy across frequency spectra. Taken together, our analyses suggest that we must ultimately strive to increase diversity and size to promote equity within genetics research.

MagicalRsq-X: A cross-cohort transferable genotype imputation quality metric

Since genotype imputation was introduced, researchers have been relying on the estimated imputation quality from imputation software to perform post-imputation quality control (QC). However, this quality estimate (denoted as Rsq) performs less well for lower-frequency variants. We recently published MagicalRsq, a machine-learning-based imputation quality calibration, which leverages additional typed markers from the same cohort and outperforms Rsq as a QC metric. In this work, we extended the original MagicalRsq to allow cross-cohort model training and named the new model MagicalRsq-X. We removed the cohort-specific estimated minor allele frequency and included linkage disequilibrium scores and recombination rates as additional features. Leveraging whole-genome sequencing data from TOPMed, specifically participants in the BioMe, JHS, WHI, and MESA studies, we performed comprehensive cross-cohort evaluations for predominantly European and African ancestral individuals based on their inferred global ancestry with the 1000 Genomes and Human Genome Diversity Project data as reference. Our results suggest MagicalRsq-X outperforms Rsq in almost every setting, with 7.3%-14.4% improvement in squared Pearson correlation with true R2, corresponding to 85-218 K variant gains. We further developed a metric to quantify the genetic distances of a target cohort relative to a reference cohort and showed that such metric largely explained the performance of MagicalRsq-X models. Finally, we found MagicalRsq-X saved up to 53 known genome-wide significant variants in one of the largest blood cell trait GWASs that would be missed using the original Rsq for QC. In conclusion, MagicalRsq-X shows superiority for post-imputation QC and benefits genetic studies by distinguishing well and poorly imputed lower-frequency variants.

A cost-effective sequencing method for genetic studies combining high-depth whole exome and low-depth whole genome

Whole genome sequencing (WGS) at high-depth (30X) allows the accurate discovery of variants in the coding and non-coding DNA regions and helps elucidate the genetic underpinnings of human health and diseases. Yet, due to the prohibitive cost of high-depth WGS, most large-scale genetic association studies use genotyping arrays or high-depth whole exome sequencing (WES). Here we propose a cost-effective method which we call "Whole Exome Genome Sequencing" (WEGS), that combines low-depth WGS and high-depth WES with up to 8 samples pooled and sequenced simultaneously (multiplexed). We experimentally assess the performance of WEGS with four different depth of coverage and sample multiplexing configurations. We show that the optimal WEGS configurations are 1.7-2.0 times cheaper than standard WES (no-plexing), 1.8-2.1 times cheaper than high-depth WGS, reach similar recall and precision rates in detecting coding variants as WES, and capture more population-specific variants in the rest of the genome that are difficult to recover when using genotype imputation methods. We apply WEGS to 862 patients with peripheral artery disease and show that it directly assesses more known disease-associated variants than a typical genotyping array and thousands of non-imputable variants per disease-associated locus.

Improving polygenic risk prediction in admixed populations by explicitly modeling ancestral-differential effects via GAUDI

Polygenic risk scores (PRS) have shown successes in clinics, but most PRS methods focus only on participants with distinct primary continental ancestry without accommodating recently-admixed individuals with mosaic continental ancestry backgrounds for different segments of their genomes. Here, we develop GAUDI, a novel penalized-regression-based method specifically designed for admixed individuals. GAUDI explicitly models ancestry-differential effects while borrowing information across segments with shared ancestry in admixed genomes. We demonstrate marked advantages of GAUDI over other methods through comprehensive simulation and real data analyses for traits with associated variants exhibiting ancestral-differential effects. Leveraging data from the Women's Health Initiative study, we show that GAUDI improves PRS prediction of white blood cell count and C-reactive protein in African Americans by > 64% compared to alternative methods, and even outperforms PRS-CSx with large European GWAS for some scenarios. We believe GAUDI will be a valuable tool to mitigate disparities in PRS performance in admixed individuals.

A rapid and reference-free imputation method for low-cost genotyping platforms

Most current genotype imputation methods are reference-based, which posed several challenges to users, such as high computational costs and reference panel inaccessibility. Thus, deep learning models are expected to create reference-free imputation methods performing with higher accuracy and shortening the running time. We proposed a imputation method using recurrent neural networks integrating with an additional discriminator network, namely GRUD. This method was applied to datasets from genotyping chips and Low-Pass Whole Genome Sequencing (LP-WGS) with the reference panels from The 1000 Genomes Project (1KGP) phase 3, the dataset of 4810 Singaporeans (SG10K), and The 1000 Vietnamese Genome Project (VN1K). Our model performed more accurately than other existing methods on multiple datasets, especially with common variants with large minor allele frequency, and shrank running time and memory usage. In summary, these results indicated that GRUD can be implemented in genomic analyses to improve the accuracy and running-time of genotype imputation.

Genotype imputation accuracy and the quality metrics of the minor ancestry in multi-ancestry reference panels

Large-scale imputation reference panels are currently available and have contributed to efficient genome-wide association studies through genotype imputation. However, whether large-size multi-ancestry or small-size population-specific reference panels are the optimal choices for under-represented populations continues to be debated. We imputed genotypes of East Asian (180k Japanese) subjects using the Trans-Omics for Precision Medicine reference panel and found that the standard imputation quality metric (Rsq) overestimated dosage r2 (squared correlation between imputed dosage and true genotype) particularly in marginal-quality bins. Variance component analysis of Rsq revealed that the increased imputed-genotype certainty (dosages closer to 0, 1 or 2) caused upward bias, indicating some systemic bias in the imputation. Through systematic simulations using different template switching rates (θ value) in the hidden Markov model, we revealed that the lower θ value increased the imputed-genotype certainty and Rsq; however, dosage r2 was insensitive to the θ value, thereby causing a deviation. In simulated reference panels with different sizes and ancestral diversities, the θ value estimates from Minimac decreased with the size of a single ancestry and increased with the ancestral diversity. Thus, Rsq could be deviated from dosage r2 for a subpopulation in the multi-ancestry panel, and the deviation represents different imputed-dosage distributions. Finally, despite the impact of the θ value, distant ancestries in the reference panel contributed only a few additional variants passing a predefined Rsq threshold. We conclude that the θ value substantially impacts the imputed dosage and the imputation quality metric value.

Imputation Accuracy Across Global Human Populations

Genotype imputation is now fundamental for genome-wide association studies but lacks fairness due to the underrepresentation of populations with non-European ancestries. The state-of-the-art imputation reference panel released by the Trans-Omics for Precision Medicine (TOPMed) initiative contains a substantial number of admixed African-ancestry and Hispanic/Latino samples to impute these populations with nearly the same accuracy as European-ancestry cohorts. However, imputation for populations primarily residing outside of North America may still fall short in performance due to persisting underrepresentation. To illustrate this point, we curated genome-wide array data from 23 publications published between 2008 to 2021. In total, we imputed over 43k individuals across 123 populations around the world. We identified a number of populations where imputation accuracy paled in comparison to that of European-ancestry populations. For instance, the mean imputation r-squared (Rsq) for 1-5% alleles in Saudi Arabians (N=1061), Vietnamese (N=1264), Thai (N=2435), and Papua New Guineans (N=776) were 0.79, 0.78, 0.76, and 0.62, respectively. In contrast, the mean Rsq ranged from 0.90 to 0.93 for comparable European populations matched in sample size and SNP content. Outside of Africa and Latin America, Rsq appeared to decrease as genetic distances to European reference increased, as predicted. Further analysis using sequencing data as ground truth suggested that imputation software may over-estimate imputation accuracy for non-European populations than European populations, suggesting further disparity between populations. Using 1496 whole genome sequenced individuals from Taiwan Biobank as a reference, we also assessed a strategy to improve imputation for non-European populations with meta-imputation, which can combine results from TOPMed with smaller population-specific reference panels. We found that meta-imputation in this design did not improve Rsq genome-wide. Taken together, our analysis suggests that with the current size of alternative reference panels, meta-imputation alone cannot improve imputation efficacy for underrepresented cohorts and we must ultimately strive to increase diversity and size to promote equity within genetics research.

Genetic Variants Associated With Hidradenitis Suppurativa

Importance

Hidradenitis suppurativa (HS) is a common and severely morbid chronic inflammatory skin disease that is reported to be highly heritable. However, the genetic understanding of HS is insufficient, and limited genome-wide association studies (GWASs) have been performed for HS, which have not identified significant risk loci.

Objective

To identify genetic variants associated with HS and to shed light on the underlying genes and genetic mechanisms.

Design, Setting, and Participants

This genetic association study recruited 753 patients with HS in the HS Program for Research and Care Excellence (HS ProCARE) at the University of North Carolina Department of Dermatology from August 2018 to July 2021. A GWAS was performed for 720 patients (after quality control) with controls from the Add Health study and then meta-analyzed with 2 large biobanks, UK Biobank (247 cases) and FinnGen (673 cases). Variants at 3 loci were tested for replication in the BioVU biobank (290 cases). Data analysis was performed from September 2021 to December 2022.

Main Outcomes and Measures

Main outcome measures are loci identified, with association of P < 1 × 10-8 considered significant.

Results

A total of 753 patients were recruited, with 720 included in the analysis. Mean (SD) age at symptom onset was 20.3 (10.57) years and at enrollment was 35.3 (13.52) years; 360 (50.0%) patients were Black, and 575 (79.7%) were female. In a meta-analysis of the 4 studies, 2 HS-associated loci were identified and replicated, with lead variants rs10512572 (P = 2.3 × 10-11) and rs17090189 (P = 2.1 × 10-8) near the SOX9 and KLF5 genes, respectively. Variants at these loci are located in enhancer regulatory elements detected in skin tissue.

Conclusions and Relevance

In this genetic association study, common variants associated with HS located near the SOX9 and KLF5 genes were associated with risk of HS. These or other nearby genes may be associated with genetic risk of disease and the development of clinical features, such as cysts, comedones, and inflammatory tunnels, that are unique to HS. New insights into disease pathogenesis related to these genes may help predict disease progression and novel treatment approaches in the future.