Polygenic risk prediction: why and when out-of-sample prediction R2 can exceed SNP-based heritability.

In polygenic score (PGS) analysis, the coefficient of determination (R2) is a key statistic to evaluate efficacy. R2 is the proportion of phenotypic variance explained by the PGS, calculated in a cohort that is independent of the genome-wide association study (GWAS) that provided estimates of allelic effect sizes. The SNP-based heritability (hSNP2, the proportion of total phenotypic variances attributable to all common SNPs) is the theoretical upper limit of the out-of-sample prediction R2. However, in real data analyses R2 has been reported to exceed hSNP2, which occurs in parallel with the observation that hSNP2 estimates tend to decline as the number of cohorts being meta-analyzed increases. Here, we quantify why and when these observations are expected. Using theory and simulation, we show that if heterogeneities in cohort-specific hSNP2 exist, or if genetic correlations between cohorts are less than one, hSNP2 estimates can decrease as the number of cohorts being meta-analyzed increases. We derive conditions when the out-of-sample prediction R2 will be greater than hSNP2 and show the validity of our derivations with real data from a binary trait (major depression) and a continuous trait (educational attainment). Our research calls for a better approach to integrating information from multiple cohorts to address issues of between-cohort heterogeneity.

Assessing polygenic risk score models for applications in populations with under-represented genomics data: an example of Vietnam.

Most polygenic risk score (PRS)models have been based on data from populations of European origins (accounting for the majority of the large genomics datasets, e.g. >78% in the UK Biobank and >85% in the GTEx project). Although several large-scale Asian biobanks were initiated (e.g. Japanese, Korean, Han Chinese biobanks), most other Asian countries have little or near-zero genomics data. To implement PRS models for under-represented populations, we explored transfer learning approaches, assuming that information from existing large datasets can compensate for the small sample size that can be feasibly obtained in developing countries, like Vietnam. Here, we benchmark 13 common PRS methods in meta-population strategy (combining individual genotype data from multiple populations) and multi-population strategy (combining summary statistics from multiple populations). Our results highlight the complementarity of different populations and the choice of methods should depend on the target population. Based on these results, we discussed a set of guidelines to help users select the best method for their datasets. We developed a robust and comprehensive software to allow for benchmarking comparisons between methods and proposed a computational framework for improving PRS performance in a dataset with a small sample size. This work is expected to inform the development of genomics applications in under-represented populations. PRSUP framework is available at: https://github.com/BiomedicalMachineLearning/VGP.

Estimation of Genetic Correlation via Linkage Disequilibrium Score Regression and Genomic Restricted Maximum Likelihood.

Genetic correlation is a key population parameter that describes the shared genetic architecture of complex traits and diseases. It can be estimated by current state-of-art methods, i.e., linkage disequilibrium score regression (LDSC) and genomic restricted maximum likelihood (GREML). The massively reduced computing burden of LDSC compared to GREML makes it an attractive tool, although the accuracy (i.e., magnitude of standard errors) of LDSC estimates has not been thoroughly studied. In simulation, we show that the accuracy of GREML is generally higher than that of LDSC. When there is genetic heterogeneity between the actual sample and reference data from which LD scores are estimated, the accuracy of LDSC decreases further. In real data analyses estimating the genetic correlation between schizophrenia (SCZ) and body mass index, we show that GREML estimates based on ∼150,000 individuals give a higher accuracy than LDSC estimates based on ∼400,000 individuals (from combined meta-data). A GREML genomic partitioning analysis reveals that the genetic correlation between SCZ and height is significantly negative for regulatory regions, which whole genome or LDSC approach has less power to detect. We conclude that LDSC estimates should be carefully interpreted as there can be uncertainty about homogeneity among combined meta-datasets. We suggest that any interesting findings from massive LDSC analysis for a large number of complex traits should be followed up, where possible, with more detailed analyses with GREML methods, even if sample sizes are lesser.

Whole-genome sequence-based genomic prediction in laying chickens with different genomic relationship matrices to account for genetic architecture.

BACKGROUND: With the availability of next-generation sequencing technologies, genomic prediction based on whole-genome sequencing (WGS) data is now feasible in animal breeding schemes and was expected to lead to higher predictive ability, since such data may contain all genomic variants including causal mutations. Our objective was to compare prediction ability with high-density (HD) array data and WGS data in a commercial brown layer line with genomic best linear unbiased prediction (GBLUP) models using various approaches to weight single nucleotide polymorphisms (SNPs). METHODS: A total of 892 chickens from a commercial brown layer line were genotyped with 336 K segregating SNPs (array data) that included 157 K genic SNPs (i.e. SNPs in or around a gene). For these individuals, genome-wide sequence information was imputed based on data from re-sequencing runs of 25 individuals, leading to 5.2 million (M) imputed SNPs (WGS data), including 2.6 M genic SNPs. De-regressed proofs (DRP) for eggshell strength, feed intake and laying rate were used as quasi-phenotypic data in genomic prediction analyses. Four weighting factors for building a trait-specific genomic relationship matrix were investigated: identical weights, -(log10 P) from genome-wide association study results, squares of SNP effects from random regression BLUP, and variable selection based weights (known as BLUP|GA). Predictive ability was measured as the correlation between DRP and direct genomic breeding values in five replications of a fivefold cross-validation. RESULTS: Averaged over the three traits, the highest predictive ability (0.366 ± 0.075) was obtained when only genic SNPs from WGS data were used. Predictive abilities with genic SNPs and all SNPs from HD array data were 0.361 ± 0.072 and 0.353 ± 0.074, respectively. Prediction with -(log10 P) or squares of SNP effects as weighting factors for building a genomic relationship matrix or BLUP|GA did not increase accuracy, compared to that with identical weights, regardless of the SNP set used. CONCLUSIONS: Our results show that little or no benefit was gained when using all imputed WGS data to perform genomic prediction compared to using HD array data regardless of the weighting factors tested. However, using only genic SNPs from WGS data had a positive effect on prediction ability.

Comparison among three variant callers and assessment of the accuracy of imputation from SNP array data to whole-genome sequence level in chicken.

BACKGROUND: The technical progress in the last decade has made it possible to sequence millions of DNA reads in a relatively short time frame. Several variant callers based on different algorithms have emerged and have made it possible to extract single nucleotide polymorphisms (SNPs) out of the whole-genome sequence. Often, only a few individuals of a population are sequenced completely and imputation is used to obtain genotypes for all sequence-based SNP loci for other individuals, which have been genotyped for a subset of SNPs using a genotyping array. METHODS: First, we compared the sets of variants detected with different variant callers, namely GATK, freebayes and SAMtools, and checked the quality of genotypes of the called variants in a set of 50 fully sequenced white and brown layers. Second, we assessed the imputation accuracy (measured as the correlation between imputed and true genotype per SNP and per individual, and genotype conflict between father-progeny pairs) when imputing from high density SNP array data to whole-genome sequence using data from around 1000 individuals from six different generations. Three different imputation programs (Minimac, FImpute and IMPUTE2) were checked in different validation scenarios. RESULTS: There were 1,741,573 SNPs detected by all three callers on the studied chromosomes 3, 6, and 28, which was 71.6 % (81.6 %, 88.0 %) of SNPs detected by GATK (SAMtools, freebayes) in total. Genotype concordance (GC) defined as the proportion of individuals whose array-derived genotypes are the same as the sequence-derived genotypes over all non-missing SNPs on the array were 0.98 (GATK), 0.97 (freebayes) and 0.98 (SAMtools). Furthermore, the percentage of variants that had high values (>0.9) for another three measures (non-reference sensitivity, non-reference genotype concordance and precision) were 90 (88, 75) for GATK (SAMtools, freebayes). With all imputation programs, correlation between original and imputed genotypes was >0.95 on average with randomly masked 1000 SNPs from the SNP array and >0.85 for a leave-one-out cross-validation within sequenced individuals. CONCLUSIONS: Performance of all variant callers studied was very good in general, particularly for GATK and SAMtools. FImpute performed slightly worse than Minimac and IMPUTE2 in terms of genotype correlation, especially for SNPs with low minor allele frequency, while it had lowest numbers in Mendelian conflicts in available father-progeny pairs. Correlations of real and imputed genotypes remained constantly high even if individuals to be imputed were several generations away from the sequenced individuals.

Integrating Single-Cell and Spatial Transcriptomics Reveals Heterogeneity of Early Pig Skin Development and a Subpopulation with Hair Placode Formation.

The dermis and epidermis, crucial structural layers of the skin, encompass appendages, hair follicles (HFs), and intricate cellular heterogeneity. However, an integrated spatiotemporal transcriptomic atlas of embryonic skin has not yet been described and would be invaluable for studying skin-related diseases in humans. Here, single-cell and spatial transcriptomic analyses are performed on skin samples of normal and hairless fetal pigs across four developmental periods. The cross-species comparison of skin cells illustrated that the pig epidermis is more representative of the human epidermis than mice epidermis. Moreover, Phenome-wide association study analysis revealed that the conserved genes between pigs and humans are strongly associated with human skin-related diseases. In the epidermis, two lineage differentiation trajectories describe hair follicle (HF) morphogenesis and epidermal development. By comparing normal and hairless fetal pigs, it is found that the hair placode (Pc), the most characteristic initial structure in HFs, arises from progenitor-like OGN+/UCHL1+ cells. These progenitors appear earlier in development than the previously described early Pc cells and exhibit abnormal proliferation and migration during differentiation in hairless pigs. The study provides a valuable resource for in-depth insights into HF development, which may serve as a key reference atlas for studying human skin disease etiology using porcine models.

Spatial transcriptomic analysis of Sonic hedgehog medulloblastoma identifies that the loss of heterogeneity and promotion of differentiation underlies the response to CDK4/6 inhibition.

BACKGROUND: Medulloblastoma (MB) is a malignant tumour of the cerebellum which can be classified into four major subgroups based on gene expression and genomic features. Single-cell transcriptome studies have defined the cellular states underlying each MB subgroup; however, the spatial organisation of these diverse cell states and how this impacts response to therapy remains to be determined. METHODS: Here, we used spatially resolved transcriptomics to define the cellular diversity within a sonic hedgehog (SHH) patient-derived model of MB and show that cells specific to a transcriptional state or spatial location are pivotal for CDK4/6 inhibitor, Palbociclib, treatment response. We integrated spatial gene expression with histological annotation and single-cell gene expression data from MB, developing an analysis strategy to spatially map cell type responses within the hybrid system of human and mouse cells and their interface within an intact brain tumour section. RESULTS: We distinguish neoplastic and non-neoplastic cells within tumours and from the surrounding cerebellar tissue, further refining pathological annotation. We identify a regional response to Palbociclib, with reduced proliferation and induced neuronal differentiation in both treated tumours. Additionally, we resolve at a cellular resolution a distinct tumour interface where the tumour contacts neighbouring mouse brain tissue consisting of abundant astrocytes and microglia and continues to proliferate despite Palbociclib treatment. CONCLUSIONS: Our data highlight the power of using spatial transcriptomics to characterise the response of a tumour to a targeted therapy and provide further insights into the molecular and cellular basis underlying the response and resistance to CDK4/6 inhibitors in SHH MB.

[Chinese Holstein Cattle effective population size estimated from whole genome linkage disequilibrium].

Effective populations size (Ne) is an important population parameter that helps to explain genetic variation, population evolution and understanding of the genetic architecture underlying complex traits. With the availability of high-dense SNP panels, more and more researches focus on estimating of Ne using linkage disequilibrium (LD) between SNPs. In this study, we estimated the effective population size from 2093 Chinese Holstein Cattle genotyped with Illumina BovineSNP50 BeadChip. After removal of individual with call rate lt; 95%, SNPs with call rate lt; 95%, minor allele frequency lt; 5% and Hardy-Weinberg Equilibrium test with Plt;0.0001, 1 968 individuals with 38 796 SNPs were remained. Eight kinds of SNP pairs with the distances 0.1, 0.2, 0.5, 1, 2, 5, 10, and 15 Mb were respectively chosen to estimate the effective population size of Chinese Holstein cattle from 4 generations ago. It is demonstrated from the results of this study that the effective population size of Chinese Holstein is decreased in the past generations, and the corresponding effective population size at ~4 generations ago is only around 45.

A Comparison of Ten Polygenic Score Methods for Psychiatric Disorders Applied Across Multiple Cohorts.

BACKGROUND: Polygenic scores (PGSs), which assess the genetic risk of individuals for a disease, are calculated as a weighted count of risk alleles identified in genome-wide association studies. PGS methods differ in which DNA variants are included and the weights assigned to them; some require an independent tuning sample to help inform these choices. PGSs are evaluated in independent target cohorts with known disease status. Variability between target cohorts is observed in applications to real data sets, which could reflect a number of factors, e.g., phenotype definition or technical factors. METHODS: The Psychiatric Genomics Consortium Working Groups for schizophrenia and major depressive disorder bring together many independently collected case-control cohorts. We used these resources (31,328 schizophrenia cases, 41,191 controls; 248,750 major depressive disorder cases, 563,184 controls) in repeated application of leave-one-cohort-out meta-analyses, each used to calculate and evaluate PGS in the left-out (target) cohort. Ten PGS methods (the baseline PC+T method and 9 methods that model genetic architecture more formally: SBLUP, LDpred2-Inf, LDpred-funct, LDpred2, Lassosum, PRS-CS, PRS-CS-auto, SBayesR, MegaPRS) were compared. RESULTS: Compared with PC+T, the other 9 methods gave higher prediction statistics, MegaPRS, LDPred2, and SBayesR significantly so, explaining up to 9.2% variance in liability for schizophrenia across 30 target cohorts, an increase of 44%. For major depressive disorder across 26 target cohorts, these statistics were 3.5% and 59%, respectively. CONCLUSIONS: Although the methods that more formally model genetic architecture have similar performance, MegaPRS, LDpred2, and SBayesR rank highest in most comparisons and are recommended in applications to psychiatric disorders.

Theoretical and empirical quantification of the accuracy of polygenic scores in ancestry divergent populations.

Polygenic scores (PGS) have been widely used to predict disease risk using variants identified from genome-wide association studies (GWAS). To date, most GWAS have been conducted in populations of European ancestry, which limits the use of GWAS-derived PGS in non-European ancestry populations. Here, we derive a theoretical model of the relative accuracy (RA) of PGS across ancestries. We show through extensive simulations that the RA of PGS based on genome-wide significant SNPs can be predicted accurately from modelling linkage disequilibrium (LD), minor allele frequencies (MAF), cross-population correlations of causal SNP effects and heritability. We find that LD and MAF differences between ancestries can explain between 70 and 80% of the loss of RA of European-based PGS in African ancestry for traits like body mass index and type 2 diabetes. Our results suggest that causal variants underlying common genetic variation identified in European ancestry GWAS are mostly shared across continents.

Detecting Genotype-Population Interaction Effects by Ancestry Principal Components.

Heterogeneity in the phenotypic mean and variance across populations is often observed for complex traits. One way to understand heterogeneous phenotypes lies in uncovering heterogeneity in genetic effects. Previous studies on genetic heterogeneity across populations were typically based on discrete groups in populations stratified by different countries or cohorts, which ignored the difference of population characteristics for the individuals within each group and resulted in loss of information. Here, we introduce a novel concept of genotype-by-population (G × P) interaction where population is defined by the first and second ancestry principal components (PCs), which are less likely to be confounded with country/cohort-specific factors. We applied a reaction norm model fitting each of 70 complex traits with significant SNP-heritability and the PCs as covariates to examine G × P interactions across diverse populations including white British and other white Europeans from the UK Biobank (N = 22,229). Our results demonstrated a significant population genetic heterogeneity for behavioral traits such as age at first sexual intercourse and academic qualification. Our approach may shed light on the latent genetic architecture of complex traits that underlies the modulation of genetic effects across different populations.

Whole-Genome Approach Discovers Novel Genetic and Nongenetic Variance Components Modulated by Lifestyle for Cardiovascular Health.

Background Both genetic and nongenetic factors can predispose individuals to cardiovascular risk. Finding ways to alter these predispositions is important for cardiovascular disease prevention. Methods and Results We used a novel whole-genome approach to estimate the genetic and nongenetic effects on-and hence their predispositions to-cardiovascular risk and determined whether they vary with respect to lifestyle factors such as physical activity, smoking, alcohol consumption, and dietary intake. We performed analyses on the ARIC (Atherosclerosis Risk in Communities) Study (N=6896-7180) and validated findings using the UKBB (UK Biobank, N=14 076-34 538). Lifestyle modulation was evident for many cardiovascular traits such as body mass index and resting heart rate. For example, alcohol consumption modulated both genetic and nongenetic effects on body mass index, whereas smoking modulated nongenetic effects on heart rate, pulse pressure, and white blood cell count. We also stratified individuals according to estimated genetic and nongenetic effects that are modulated by lifestyle factors and showed distinct phenotype-lifestyle relationships across the stratified groups. Finally, we showed that neglecting lifestyle modulations of cardiovascular traits would on average reduce single nucleotide polymorphism heritability estimates of these traits by a small yet significant amount, primarily owing to the overestimation of residual variance. Conclusions Lifestyle changes are relevant to cardiovascular disease prevention. Individual differences in the genetic and nongenetic effects that are modulated by lifestyle factors, as shown by the stratified group analyses, implies a need for personalized lifestyle interventions. In addition, single nucleotide polymorphism-based heritability of cardiovascular traits without accounting for lifestyle modulations could be underestimated.

The genetic relationship between female reproductive traits and six psychiatric disorders.

Female reproductive behaviours have important implications for evolutionary fitness and health of offspring. Here we used the second release of UK Biobank data (N = 220,685) to evaluate the association between five female reproductive traits and polygenic risk scores (PRS) projected from genome-wide association study summary statistics of six psychiatric disorders (N = 429,178). We found that the PRS of attention-deficit/hyperactivity disorder (ADHD) were strongly associated with age at first birth (AFB) (genetic correlation of -0.68 ± 0.03), age at first sexual intercourse (AFS) (-0.56 ± 0.03), number of live births (NLB) (0.36 ± 0.04) and age at menopause (-0.27 ± 0.04). There were also robustly significant associations between the PRS of eating disorder (ED) and AFB (0.35 ± 0.06), ED and AFS (0.19 ± 0.06), major depressive disorder (MDD) and AFB (-0.27 ± 0.07), MDD and AFS (-0.27 ± 0.03) and schizophrenia and AFS (-0.10 ± 0.03). These associations were mostly explained by pleiotropic effects and there was little evidence of causal relationships. Our findings can potentially help improve reproductive health in women, hence better child outcomes. Our findings also lend partial support to the evolutionary hypothesis that causal mutations underlying psychiatric disorders have positive effects on reproductive success.

Genotype-covariate correlation and interaction disentangled by a whole-genome multivariate reaction norm model.

The genomics era has brought useful tools to dissect the genetic architecture of complex traits. Here we propose a multivariate reaction norm model (MRNM) to tackle genotype-covariate (G-C) correlation and interaction problems. We apply MRNM to the UK Biobank data in analysis of body mass index using smoking quantity as a covariate, finding a highly significant G-C correlation, but only weak evidence for G-C interaction. In contrast, G-C interaction estimates are inflated in existing methods. It is also notable that there is significant heterogeneity in the estimated residual variances (i.e., variances not attributable to factors in the model) across different covariate levels, i.e., residual-covariate (R-C) interaction. We also show that the residual variances estimated by standard additive models can be inflated in the presence of G-C and/or R-C interactions. We conclude that it is essential to correctly account for both interaction and correlation in complex trait analyses.

Age at first birth in women is genetically associated with increased risk of schizophrenia.

Previous studies have shown an increased risk for mental health problems in children born to both younger and older parents compared to children of average-aged parents. We previously used a novel design to reveal a latent mechanism of genetic association between schizophrenia and age at first birth in women (AFB). Here, we use independent data from the UK Biobank (N = 38,892) to replicate the finding of an association between predicted genetic risk of schizophrenia and AFB in women, and to estimate the genetic correlation between schizophrenia and AFB in women stratified into younger and older groups. We find evidence for an association between predicted genetic risk of schizophrenia and AFB in women (P-value = 1.12E-05), and we show genetic heterogeneity between younger and older AFB groups (P-value = 3.45E-03). The genetic correlation between schizophrenia and AFB in the younger AFB group is -0.16 (SE = 0.04) while that between schizophrenia and AFB in the older AFB group is 0.14 (SE = 0.08). Our results suggest that early, and perhaps also late, age at first birth in women is associated with increased genetic risk for schizophrenia in the UK Biobank sample. These findings contribute new insights into factors contributing to the complex bio-social risk architecture underpinning the association between parental age and offspring mental health.