Bias Characterization in Probabilistic Genotype Data and Improved Signal Detection with Multiple Imputation

TISSUE: uncertainty-calibrated prediction of single-cell spatial transcriptomics improves downstream analyses

Whole-transcriptome spatial profiling of genes at single-cell resolution remains a challenge. To address this limitation, spatial gene expression prediction methods have been developed to infer the spatial expression of unmeasured transcripts, but the quality of these predictions can vary greatly. Here we present Transcript Imputation with Spatial Single-cell Uncertainty Estimation (TISSUE) as a general framework for estimating uncertainty for spatial gene expression predictions and providing uncertainty-aware methods for downstream inference. Leveraging conformal inference, TISSUE provides well-calibrated prediction intervals for predicted expression values across 11 benchmark datasets. Moreover, it consistently reduces the false discovery rate for differential gene expression analysis, improves clustering and visualization of predicted spatial transcriptomics and improves the performance of supervised learning models trained on predicted gene expression profiles. Applying TISSUE to a MERFISH spatial transcriptomics dataset of the adult mouse subventricular zone, we identified subtypes within the neural stem cell lineage and developed subtype-specific regional classifiers.

TISSUE: uncertainty-calibrated prediction of single-cell spatial transcriptomics improves downstream analyses

Whole-transcriptome spatial profiling of genes at single-cell resolution remains a challenge. To address this limitation, spatial gene expression prediction methods have been developed to infer the spatial expression of unmeasured transcripts, but the quality of these predictions can vary greatly. Here we present TISSUE (Transcript Imputation with Spatial Single-cell Uncertainty Estimation) as a general framework for estimating uncertainty for spatial gene expression predictions and providing uncertainty-aware methods for downstream inference. Across eleven benchmark datasets, TISSUE provides well-calibrated prediction intervals for predicted expression values. Moreover it consistently reduces false discovery rates for differential gene expression analysis, improves clustering and visualization of predicted spatial transcriptomics, and improves the performance of supervised learning models trained on predicted gene expression profiles. Applying TISSUE to a MERFISH spatial transcriptomics dataset of the adult mouse subventricular zone, we identified subtypes within the neural stem cell lineage and developed subtype-specific regional classifiers. TISSUE is publicly available as a flexible wrapper method for existing spatial gene expression prediction methods to assist researchers with implementing uncertainty-aware analyses of spatial transcriptomics data.

Comparison of multiple imputation and other methods for the analysis of imputed genotypes

BACKGROUND

Analysis of imputed genotypes is an important and routine component of genome-wide association studies and the increasing size of imputation reference panels has facilitated the ability to impute and test low-frequency variants for associations. In the context of genotype imputation, the true genotype is unknown and genotypes are inferred with uncertainty using statistical models. Here, we present a novel method for integrating imputation uncertainty into statistical association tests using a fully conditional multiple imputation (MI) approach which is implemented using the Substantive Model Compatible Fully Conditional Specification (SMCFCS). We compared the performance of this method to an unconditional MI and two additional approaches that have been shown to demonstrate excellent performance: regression with dosages and a mixture of regression models (MRM).

RESULTS

Our simulations considered a range of allele frequencies and imputation qualities based on data from the UK Biobank. We found that the unconditional MI was computationally costly and overly conservative across a wide range of settings. Analyzing data with Dosage, MRM, or MI SMCFCS resulted in greater power, including for low frequency variants, compared to unconditional MI while effectively controlling type I error rates. MRM andl MI SMCFCS are both more computationally intensive then using Dosage.

CONCLUSIONS

The unconditional MI approach for association testing is overly conservative and we do not recommend its use in the context of imputed genotypes. Given its performance, speed, and ease of implementation, we recommend using Dosage for imputed genotypes with MAF [Formula: see text] 0.001 and Rsq [Formula: see text] 0.3.

Identifying and correcting for misspecifications in GWAS summary statistics and polygenic scores

Publicly available genome-wide association studies (GWAS) summary statistics exhibit uneven quality, which can impact the validity of follow-up analyses. First, we present an overview of possible misspecifications that come with GWAS summary statistics. Then, in both simulations and real-data analyses, we show that additional information such as imputation INFO scores, allele frequencies, and per-variant sample sizes in GWAS summary statistics can be used to detect possible issues and correct for misspecifications in the GWAS summary statistics. One important motivation for us is to improve the predictive performance of polygenic scores built from these summary statistics. Unfortunately, owing to the lack of reporting standards for GWAS summary statistics, this additional information is not systematically reported. We also show that using well-matched linkage disequilibrium (LD) references can improve model fit and translate into more accurate prediction. Finally, we discuss how to make polygenic score methods such as lassosum and LDpred2 more robust to these misspecifications to improve their predictive power.

Dietary, Cultural and Pathogens-Related Selective Pressures Shaped Differential Adaptive Evolution Among Native Mexican Populations

Native American genetic ancestry has been remarkably implicated with increased risk of diverse health issues in several Mexican populations, especially in relation to the dramatic changes in environmental, dietary, and cultural settings they have recently undergone. In particular, the effects of these ecological transitions and Westernization of lifestyles have been investigated so far predominantly on Mestizo individuals. Nevertheless, indigenous groups, rather than admixed Mexicans, have plausibly retained the highest proportions of genetic components shaped by natural selection in response to the ancient milieu experienced by Mexican ancestors during their pre-Columbian evolutionary history. These formerly adaptive variants have the potential to represent the genetic determinants of some biological traits that are peculiar to Mexican people, as well as a reservoir of loci with possible biomedical relevance. To test such a hypothesis, we used genome-wide genotype data to infer the unique adaptive evolution of Native Mexican groups selected as reasonable descendants of the main pre-Columbian Mexican civilizations. A combination of haplotype-based and gene-network analyses enabled us to detect genomic signatures ascribable to polygenic adaptive traits plausibly evolved by the main genetic clusters of Mexican indigenous populations to cope with local environmental and/or cultural conditions. Some of these adaptations were found to play a role in modulating the susceptibility/resistance of these groups to certain pathological conditions, thus providing new evidence that diverse selective pressures have contributed to shape the current biological and disease-risk patterns of present-day Native and Mestizo Mexican populations.

False positive findings during genome-wide association studies with imputation: Influence of allele frequency and imputation accuracy

Genotype imputation is widely used in genetic studies to boost the power of GWAS, to combine multiple studies for meta-analysis and to perform fine mapping. With advances of imputation tools and large reference panels, genotype imputation has become mature and accurate. However, the uncertain nature of imputed genotypes can cause bias in the downstream analysis. Many studies have compared the performance of popular imputation approaches, but few investigated bias characteristics of downstream association analyses. Herein, we showed that the imputation accuracy is diminished if the real genotypes contain minor alleles. Although these genotypes are less common, which is particularly true for loci with low minor allele frequency, a large discordance between imputed and observed genotypes significantly inflated the association results, especially in data with a large portion of uncertain SNPs. The significant discordance of p-values happened as the p-value approached 0 or the imputation quality was poor. Although elimination of poorly imputed SNPs can remove false positive (FP) SNPs, it sacrificed, sometimes, more than 80% true positive (TP) SNPs. For top ranked SNPs, removing variants with moderate imputation quality cannot reduce the proportion of FP SNPs, and increasing sample size in reference panels did not greatly benefit the results as well. Additionally, samples with a balanced ratio between cases and controls can dramatically improve the number of TP SNPs observed in the imputation based GWAS. These results raise concerns about results from analysis of association studies when rare variants are studied, particularly when case-control studies are unbalanced.

Accurate Imputation of Untyped Variants from Deep Sequencing Data

The quality, statistical power, and resolution of genome-wide association studies (GWAS) are largely dependent on the comprehensiveness of genotypic data. Over the last few years, despite the constant decrease in the price of sequencing, whole-genome sequencing (WGS) of association panels comprising a large number of samples remains cost-prohibitive. Therefore, most GWAS populations are still genotyped using low-coverage genotyping methods resulting in incomplete datasets. Imputation of untyped variants is a powerful method to maximize the number of SNPs identified in study samples, it increases the power and resolution of GWAS and allows to integrate genotyping datasets obtained from various sources. Here, we describe the key concepts underlying imputation of untyped variants, including the architecture of reference panels, and review some of the associated challenges and how these can be addressed. We also discuss the need and available methods to rigorously assess the accuracy of imputed data prior to their use in any genetic study.

Comparisons of improved genomic predictions generated by different imputation methods for genotyping by sequencing data in livestock populations

Background

Genotyping by sequencing (GBS) still has problems with missing genotypes. Imputation is important for using GBS for genomic predictions, especially for low depths, due to the large number of missing genotypes. Minor allele frequency (MAF) is widely used as a marker data editing criteria for genomic predictions. In this study, three imputation methods (Beagle, IMPUTE2 and FImpute software) based on four MAF editing criteria were investigated with regard to imputation accuracy of missing genotypes and accuracy of genomic predictions, based on simulated data of livestock population.

Results

Four MAFs (no MAF limit, MAF ≥ 0.001, MAF ≥ 0.01 and MAF ≥ 0.03) were used for editing marker data before imputation. Beagle, IMPUTE2 and FImpute software were applied to impute the original GBS. Additionally, IMPUTE2 also imputed the expected genotype dosage after genotype correction (GcIM). The reliability of genomic predictions was calculated using GBS and imputed GBS data. The results showed that imputation accuracies were the same for the three imputation methods, except for the data of sequencing read depth (depth) = 2, where FImpute had a slightly lower imputation accuracy than Beagle and IMPUTE2. GcIM was observed to be the best for all of the imputations at depth = 4, 5 and 10, but the worst for depth = 2. For genomic prediction, retaining more SNPs with no MAF limit resulted in higher reliability. As the depth increased to 10, the prediction reliabilities approached those using true genotypes in the GBS loci. Beagle and IMPUTE2 had the largest increases in prediction reliability of 5 percentage points, and FImpute gained 3 percentage points at depth = 2. The best prediction was observed at depth = 4, 5 and 10 using GcIM, but the worst prediction was also observed using GcIM at depth = 2.

Conclusions

The current study showed that imputation accuracies were relatively low for GBS with low depths and high for GBS with high depths. Imputation resulted in larger gains in the reliability of genomic predictions for GBS with lower depths. These results suggest that the application of IMPUTE2, based on a corrected GBS (GcIM) to improve genomic predictions for higher depths, and FImpute software could be a good alternative for routine imputation.

Making the Most of Clumping and Thresholding for Polygenic Scores

Polygenic prediction has the potential to contribute to precision medicine. Clumping and thresholding (C+T) is a widely used method to derive polygenic scores. When using C+T, several p value thresholds are tested to maximize predictive ability of the derived polygenic scores. Along with this p value threshold, we propose to tune three other hyper-parameters for C+T. We implement an efficient way to derive thousands of different C+T scores corresponding to a grid over four hyper-parameters. For example, it takes a few hours to derive 123K different C+T scores for 300K individuals and 1M variants using 16 physical cores. We find that optimizing over these four hyper-parameters improves the predictive performance of C+T in both simulations and real data applications as compared to tuning only the p value threshold. A particularly large increase can be noted when predicting depression status, from an AUC of 0.557 (95% CI: [0.544-0.569]) when tuning only the p value threshold to an AUC of 0.592 (95% CI: [0.580-0.604]) when tuning all four hyper-parameters we propose for C+T. We further propose stacked clumping and thresholding (SCT), a polygenic score that results from stacking all derived C+T scores. Instead of choosing one set of hyper-parameters that maximizes prediction in some training set, SCT learns an optimal linear combination of all C+T scores by using an efficient penalized regression. We apply SCT to eight different case-control diseases in the UK biobank data and find that SCT substantially improves prediction accuracy with an average AUC increase of 0.035 over standard C+T.

Genetics of Human Longevity Within an Eco-Evolutionary Nature-Nurture Framework

Human longevity is a complex trait, and to disentangle its basis has a great theoretical and practical consequences for biomedicine. The genetics of human longevity is still poorly understood despite several investigations that used different strategies and protocols. Here, we argue that such rather disappointing harvest is largely because of the extraordinary complexity of the longevity phenotype in humans. The capability to reach the extreme decades of human lifespan seems to be the result of an intriguing mixture of gene-environment interactions. Accordingly, the genetics of human longevity is here described as a highly context-dependent phenomenon, within a new integrated, ecological, and evolutionary perspective, and is presented as a dynamic process, both historically and individually. The available literature has been scrutinized within this perspective, paying particular attention to factors (sex, individual biography, family, population ancestry, social structure, economic status, and education, among others) that have been relatively neglected. The strength and limitations of the most powerful and used tools, such as genome-wide association study and whole-genome sequencing, have been discussed, focusing on prominently emerged genes and regions, such as apolipoprotein E, Forkhead box O3, interleukin 6, insulin-like growth factor-1, chromosome 9p21, 5q33.3, and somatic mutations among others. The major results of this approach suggest that (1) the genetics of longevity is highly population specific; (2) small-effect alleles, pleiotropy, and the complex allele timing likely play a major role; (3) genetic risk factors are age specific and need to be integrated in the light of the geroscience perspective; (4) a close relationship between genetics of longevity and genetics of age-related diseases (especially cardiovascular diseases) do exist. Finally, the urgent need of a global approach to the largely unexplored interactions between the 3 genetics of human body, that is, nuclear, mitochondrial, and microbiomes, is stressed. We surmise that the comprehensive approach here presented will help in increasing the above-mentioned harvest.

Efficient analysis of large-scale genome-wide data with two R packages: bigstatsr and bigsnpr

Motivation

Genome-wide datasets produced for association studies have dramatically increased in size over the past few years, with modern datasets commonly including millions of variants measured in dozens of thousands of individuals. This increase in data size is a major challenge severely slowing down genomic analyses, leading to some software becoming obsolete and researchers having limited access to diverse analysis tools.

Results

Here we present two R packages, bigstatsr and bigsnpr, allowing for the analysis of large scale genomic data to be performed within R. To address large data size, the packages use memory-mapping for accessing data matrices stored on disk instead of in RAM. To perform data pre-processing and data analysis, the packages integrate most of the tools that are commonly used, either through transparent system calls to existing software, or through updated or improved implementation of existing methods. In particular, the packages implement fast and accurate computations of principal component analysis and association studies, functions to remove single nucleotide polymorphisms in linkage disequilibrium and algorithms to learn polygenic risk scores on millions of single nucleotide polymorphisms. We illustrate applications of the two R packages by analyzing a case–control genomic dataset for celiac disease, performing an association study and computing polygenic risk scores. Finally, we demonstrate the scalability of the R packages by analyzing a simulated genome-wide dataset including 500 000 individuals and 1 million markers on a single desktop computer.

Availability and implementation

https://privefl.github.io/bigstatsr/ and https://privefl.github.io/bigsnpr/.

Supplementary information

Supplementary data are available at Bioinformatics online.

Improving genomic predictions by correction of genotypes from genotyping by sequencing in livestock populations

Background

Genotyping by sequencing (GBS) is a robust method to genotype markers. Many factors can influence the genotyping quality. One is that heterozygous genotypes could be wrongly genotyped as homozygotes, dependent on the genotyping depths. In this study, a method correcting this type of genotyping error was demonstrated. The efficiency of this correction method and its effect on genomic prediction were assessed using simulated data of livestock populations.

Results

Chip array (Chip) and four depths of GBS data was simulated. After quality control (call rate ≥ 0.8 and MAF ≥ 0.01), the remaining number of Chip and GBS SNPs were both approximately 7,000, averaged over 10 replicates. GBS genotypes were corrected with the proposed method. The reliability of genomic prediction was calculated using GBS, corrected GBS (GBSc), true genotypes for the GBS loci (GBSr) and Chip data. The results showed that GBSc had higher rates of correct genotype calls and higher correlations with true genotypes than GBS. For genomic prediction, using Chip data resulted in the highest reliability. As the depth increased to 10, the prediction reliabilities using GBS and GBSc data approached those using true GBS data. The reliabilities of genomic prediction using GBSc data were 0.604, 0.672, 0.684 and 0.704 after genomic correction, with the improved values of 0.013, 0.009, 0.006 and 0.001 at depth = 2, 4, 5 and 10, respectively.

Conclusions

The current study showed that a correction method for GBS data increased the genotype accuracies and, consequently, improved genomic predictions. These results suggest that a correction of GBS genotype is necessary, especially for the GBS data with low depths.

Electronic supplementary material

The online version of this article (10.1186/s40104-019-0315-z) contains supplementary material, which is available to authorized users.