Modeling Linkage Disequilibrium Increases Accuracy of Polygenic Risk Scores

Genomic landscape of cancer in racially and ethnically diverse populations

Cancer incidence and mortality rates can vary widely among different racial and ethnic groups, attributed to a complex interplay of genetic, environmental and social factors. Recently, substantial progress has been made in investigating hereditary genetic risk factors and in characterizing tumour genomes. However, most research has been conducted in individuals of European ancestries and, increasingly, in individuals of Asian ancestries. The study of germline and somatic genetics in cancer across racial and ethnic groups using omics technologies offers opportunities to identify similarities and differences in both heritable traits and the molecular features of cancer genomes. An improved understanding of population-specific cancer genomics, as well as translation of those findings across populations, will help reduce cancer disparities and ensure that personalized medicine and public health approaches are equitable across racial and ethnic groups.

The accuracy of polygenic score models for BMI and Type II diabetes in the Native Hawaiian population

Polygenic scores (PGS) are promising in stratifying individuals based on the genetic susceptibility to complex diseases or traits. However, the accuracy of PGS models, typically trained in European- or East Asian-ancestry populations, tend to perform poorly in other ethnic minority populations and their accuracies have not been evaluated for Native Hawaiians. In particular, for body mass index (BMI) and type-2 diabetes (T2D), Polynesian-ancestry individuals such as Native Hawaiians or Samoans exhibit varied distribution from other continental populations, but are understudied, particularly in the context of PGS. Using BMI and T2D as examples of metabolic traits of importance to Polynesian populations (along with height as a comparison of a similarly highly polygenic trait), here we examine the prediction accuracies of PGS models in a large Native Hawaiian sample from the Multiethnic Cohort with up to 5300 individuals. We find evidence of lowered prediction accuracies for the PGS models in some cases, particularly for height. We also find that using the Native Hawaiian samples as an optimization cohort during training does not consistently improve PGS performance. Moreover, even the best-performing PGS models among Native Hawaiians have lowered prediction accuracy among the subset of individuals most enriched with Polynesian ancestry. Our findings indicate that factors such as admixture histories, sample size, and diversity in GWAS can influence PGS performance for complex traits among Native Hawaiian samples. This study provides an initial survey of PGS performance among Native Hawaiians and exposes the current gaps and challenges associated with improving polygenic prediction models for underrepresented minority populations.

Characterizing selection on complex traits through conditional frequency spectra

Natural selection on complex traits is difficult to study in part due to the ascertainment inherent to genome-wide association studies (GWAS). The power to detect a trait-associated variant in GWAS is a function of its frequency and effect size - but for traits under selection, the effect size of a variant determines the strength of selection against it, constraining its frequency. Recognizing the biases inherent to GWAS ascertainment, we propose studying the joint distribution of allele frequencies across populations, conditional on the frequencies in the GWAS cohort. Before considering these conditional frequency spectra, we first characterized the impact of selection and non-equilibrium demography on allele frequency dynamics forwards and backwards in time. We then used these results to understand conditional frequency spectra under realistic human demography. Finally, we investigated empirical conditional frequency spectra for GWAS variants associated with 106 complex traits, finding compelling evidence for either stabilizing or purifying selection. Our results provide insights into polygenic score portability and other properties of variants ascertained with GWAS, highlighting the utility of conditional frequency spectra.

Fast analysis of biobank-size data and meta-analysis using the BGLR R-package

Analyzing human genomic data from biobanks and large-scale genetic evaluations often requires fitting models with a sample size exceeding the number of DNA markers used (n>p). For instance, developing polygenic scores for humans and genomic prediction for genetic evaluations of agricultural species may require fitting models involving a few thousand SNPs using data with hundreds of thousands of samples. In such cases, computations based on sufficient statistics are more efficient than those based on individual genotype-phenotype data. Additionally, software that admits sufficient statistics as inputs can be used to analyze data from multiple sources jointly without the need to share individual genotype-phenotype data. Therefore, we developed functionality within the BGLR R-package that generates posterior samples for Bayesian shrinkage and variable selection models from sufficient statistics. In this article, we present an overview of the new methods incorporated in the BGLR R-package, demonstrate the use of the new software through simple examples, provide several computational benchmarks, and present a real-data example using data from the UK-Biobank, All of Us, and the Hispanic Community Health Study/Study of Latinos cohort demonstrating how a joint analysis from multiple cohorts can be implemented without sharing individual genotype-phenotype data, and how a combined analysis can improve the prediction accuracy of polygenic scores for Hispanics-a group severely under-represented in genome-wide association studies data.

Associations of polygenic risk score, environmental factors, and their interactions with the risk of schizophrenia spectrum disorders

BACKGROUND

Emerging evidence indicates that gene-environment interactions (GEIs) are important underlying mechanisms for the development of schizophrenia (SZ). We investigated the associations of polygenic risk score for SZ (PRS-SZ), environmental measures, and their interactions with case-control status and clinical phenotypes among patients with schizophrenia spectrum disorders (SSDs).

METHODS

The PRS-SZ for 717 SSD patients and 356 healthy controls (HCs) were calculated using the LDpred model. The Korea-Polyenvironmental Risk Score-I (K-PERS-I) and Early Trauma Inventory-Self Report (ETI-SR) were utilized as environmental measures. Logistic and linear regression analyses were performed to identify the associations of PRS-SZ and two environmental measures with case-control status and clinical phenotypes.

RESULTS

The PRS-SZ explained 8.7% of SZ risk. We found greater associations of PRS-SZ and total scores of the K-PERS-I with case-control status compared to the ETI-SR total score. A significant additive interaction was found between PRS-SZ and K-PERS-I. With the subdomains of the K-PERS-I and ETI-SR, we identified significant multiplicative or additive interactions of PRS-SZ and parental socioeconomic status (pSES), childhood adversity, and recent life events in association with case-control status. For clinical phenotypes, significant interactions were observed between PRS-SZ and the ETI-SR total score for negative-self and between PRS-SZ and obstetric complications within the K-PERS-I for negative-others.

CONCLUSIONS

Our findings suggest that the use of aggregate scores for genetic and environmental measures, PRS-SZ and K-PERS-I, can more accurately predict case-control status, and specific environmental measures may be more suitable for the exploration of GEIs.

Enhancing nonlinear transcriptome- and proteome-wide association studies via trait imputation with applications to Alzheimer's disease

Genome-wide association studies (GWAS) performed on large cohort and biobank datasets have identified many genetic loci associated with Alzheimer's disease (AD). However, the younger demographic of biobank participants relative to the typical age of late-onset AD has resulted in an insufficient number of AD cases, limiting the statistical power of GWAS and any downstream analyses. To mitigate this limitation, several trait imputation methods have been proposed to impute the expected future AD status of individuals who may not have yet developed the disease. This paper explores the use of imputed AD status in nonlinear transcriptome/proteome-wide association studies (TWAS/PWAS) to identify genes and proteins whose genetically regulated expression is associated with AD risk. In particular, we considered the TWAS/PWAS method DeLIVR, which utilizes deep learning to model the nonlinear effects of expression on disease. We trained transcriptome and proteome imputation models for DeLIVR on data from the Genotype-Tissue Expression (GTEx) Project and the UK Biobank (UKB), respectively, with imputed AD status in UKB participants as the outcome. Next, we performed hypothesis testing for the DeLIVR models using clinically diagnosed AD cases from the Alzheimer's Disease Sequencing Project (ADSP). Our results demonstrate that nonlinear TWAS/PWAS trained with imputed AD outcomes successfully identifies known and putative AD risk genes and proteins. Notably, we found that training with imputed outcomes can increase statistical power without inflating false positives, enabling the discovery of molecular exposures with potentially nonlinear effects on neurodegeneration.

Incorporating multiple functional annotations to improve polygenic risk prediction accuracy

We present OmniPRS, a scalable biobank-scale framework that improves genetic risk prediction for complex traits by integrating genome-wide association study (GWAS) summary statistics and functional annotations. It employs a mixed model incorporating tissue-specific genetic variance components from annotations to re-estimate single-nucleotide polymorphism (SNP) effects and constructs tissue-specific polygenic risk scores (PRSs) and aggregates them into the final OmniPRS. Our experiments, encompassing 135 simulation scenarios and 11 representative traits, demonstrate that OmniPRS is flexible and robust, delivering efficient and accurate predictions comparable to ten leading PRS methods. For quantitative (binary) traits, OmniPRS achieved an average improvement of 52.31% (19.83%) versus the clumping and thresholding (C+T) method, 3.92% (1.31%) versus the annotation-integrated PRSs (LDpred-funct), and 8.44% (2.27%) versus the Bayesian-based PRSs (PRScs). Notably, it achieved 35× faster computation than the PRScs. This rapid, precise framework enables efficient polygenic risk scoring with multi-annotation integration for large-scale genomic studies.

Intergenerational transmission of complex traits and the offspring methylome

The genetic makeup of parents can directly or indirectly affect their offspring phenome through genetic transmission or via the environment that is influenced by parental heritable traits. Our understanding of the mechanisms by which indirect genetic effects operate is limited. Here, we hypothesize that one mechanism is via the offspring methylome. To test this hypothesis, polygenic scores (PGSs) for schizophrenia, smoking initiation, educational attainment (EA), social deprivation, body mass index (BMI), and height were analyzed in a cohort of 1528 offspring and their parents (51.5% boys, mean [SD] age = 10 [2.8] years). We modelled parent and offspring PGSs on offspring buccal-DNA methylation, accounting for the own PGS of offspring, and found significant associations between parental PGSs for schizophrenia, EA, BMI, and height, and offspring buccal methylation sites, comprising 16, 2, 1, and 6 sites, respectively (alpha = 2.7 × 10-5). More DNA methylation sites were associated with maternal than paternal PGSs, possibly reflecting the maternal pre- and periconceptional environment or stronger maternal involvement in shaping the offspring's environment during early childhood.

Polygenic Risk Score and Lung Adenocarcinoma Risk Among Never-Smokers by EGFR Mutation Status: A Brief Report

We assessed the association between a genome-wide polygenic risk score (PRS) developed for lung adenocarcinoma (LUAD) risk and mutation on the EGFR gene in 998 East Asian never-smoking female LUAD cases (518 EGFR-positive; 480 EGFR-negative) and 4544 never-smoking controls using case-case and multinomial regression analyses. We found that the PRS was more strongly associated with EGFR-positive LUAD compared with EGFR-negative LUAD, where the association between the fourth quartile of the PRS and EGFR-positive LUAD (odds ratio = 8.63, 95% confidence interval: 5.67-13.14) was significantly higher than the association between the fourth quartile of the PRS with EGFR-negative LUAD (odds ratio = 3.50, 95% confidence interval: 2.44-5.00) (p-heterogeneity = 3.66 × 10-3). Our findings suggest that germline genetic susceptibility may be differentially associated with LUAD in never-smoking female East Asian patients depending on the cancer's mutation status, which may have important public health and clinical implications.

Diagnostic testing in the genetically complex age-related macular degeneration

Age-related macular degeneration (AMD) is a leading cause of visual impairment with the risk of developing the disease influenced by a combination of genetic and environmental factors. With the recent expansion of treatment options, enhancing diagnostic accuracy and improving access to treatment are increasingly becoming the focus of interest. By using data from genome-wide association studies (GWAS) to generate polygenic risk scores (PRS), an assessment of an individual's genetic risk for AMD is feasible. While the predictive accuracy of the AMD-PRS is most robust for individuals at very high genetic risk, genetic diagnostic testing is warranted due to the large number of affected individuals resulting from the high prevalence of AMD. Early genetic confirmation of AMD-related pathology can facilitate timely treatment initiation, potentially improving patient outcomes.

Unsupervised Ensemble Learning for Efficient Integration of Pre-trained Polygenic Risk Scores

The growing availability of pre-trained polygenic risk score (PRS) models has enabled their integration into real-world applications, reducing the need for extensive data labeling, training, and calibration. However, selecting the most suitable PRS model for a specific target population remains challenging, due to issues such as limited transferability, data heterogeneity, and the scarcity of observed phenotype in real-world settings. Ensemble learning offers a promising avenue to enhance the predictive accuracy of genetic risk assessments, but most existing methods often rely on observed phenotype data or additional genome-wide association studies (GWAS) from the target population to optimize ensemble weights, limiting their utility in real-time implementation. Here, we present the UNSupervised enSemble PRS (UNSemblePRS), an unsupervised ensemble learning framework, that combines pre-trained PRS models without requiring phenotype data or summaries from the target population. Unlike traditional supervised approaches, UNSemblePRS aggregates models based on prediction concordance across a curated subset of candidate PRS models. We evaluated UNSemblePRS using both continuous and binary traits in the All of Us database, demonstrating its scalability and robust performance across diverse populations. These results underscore UNSemblePRS as an accessible tool for integrating PRS models into real-world contexts, offering broad applicability as the availability of PRS models continues to expand.

Efficient blockLASSO for polygenic scores with applications to all of us and UK Biobank

We develop a "block" LASSO (blockLASSO) approach for training polygenic scores (PGS) and demonstrate its use in All of Us (AoU) and the UK Biobank (UKB). blockLASSO utilizes the approximate block diagonal structure (due to chromosomal partition of the genome) of linkage disequilibrium (LD). The new implementation can be used for exploratory and methods research where repeated PGS training is necessary and expensive. For 11 different phenotypes, in two different biobanks, and across 5 different ancestry groups (African, American, East Asian, European, and South Asian) - we demonstrate that blockLASSO is generally as effective for training PGS as a (global) LASSO. Previous work has shown penalized regression methods produce competitive PGS to alternative approaches. It has been shown that some phenotypes are more/less polygenic than others. Using sparse algorithms, an accurate PGS can be trained for type 1 diabetes (T1D) using ∼ 100 single nucleotide variants (SNVs), but a PGS for body mass index (BMI) would need more than 10k SNVs. blockLASSO produces similar PGS for phenotypes while training with just a fraction of the variants per block. Within AoU (using only genetic information) block PGS for T1D reaches an AUC of 0 . 63 ± 0.02 and for BMI a correlation of 0 . 21 ± 0.01 , whereas a global LASSO approach which finds for T1D an AUC 0 . 65 ± 0.03 and BMI a correlation 0 . 19 ± 0.03 . This new block approach is more computationally efficient and scalable than naive global machine learning approaches and makes it ideal for exploratory methods investigations based on penalized regression.

Estimating Disorder Probability Based on Polygenic Prediction Using the BPC Approach

Polygenic Scores (PGSs) summarize an individual's genetic propensity for a given trait in a single value based on SNP effect sizes derived from Genome-Wide Association Study (GWAS) results. Methods have been developed that apply Bayesian approaches to improve the prediction accuracy of PGSs through optimization of estimated effect sizes. While these methods are generally well-calibrated for continuous traits (implying the predicted values are, on average, equal to the true trait values), they are not well-calibrated for binary disorder traits in ascertained samles. This is a problem because well-calibrated PGSs are needed to reliably compute the absolute disorder probability for an individual to facilitate future clinical implementation. Here, we introduce the Bayesian polygenic score Probability Conversion (BPC) approach, which computes an individual's predicted disorder probability using GWAS summary statistics, an existing Bayesian PGS method (e.g., PRScs, SBayesR), the individual's genotype data, and a prior disorder probability (which can be specified flexibly, based on e.g., literature, small reference samples, or prior elicitation). The BPC approach transforms the PGS to its underlying liability scale, computes the variances of the PGS in cases and controls, and applies Bayes' Theorem to compute the absolute disorder probability; it is practical in its application as it does not require a tuning sample with both genotype and phenotype data. We applied the BPC approach to extensive simulated data and empirical data of nine disorders. The BPC approach yielded well-calibrated results that were consistently better than the results of another recently published approach.

Brain and blood transcriptome-wide association studies identify five novel genes associated with Alzheimer's disease

BackgroundGenome-wide association studies (GWAS) have identified numerous genetic variants associated with Alzheimer's disease (AD), but their functional implications remain unclear. Transcriptome-wide association studies (TWAS) offer enhanced statistical power by analyzing genetic associations at the gene level rather than at the variant level, enabling assessment of how genetically-regulated gene expression influences AD risk. However, previous AD-TWAS have been limited by small expression quantitative trait loci (eQTL) reference datasets or reliance on AD-by-proxy phenotypes.ObjectiveTo perform the most powerful AD-TWAS to date using summary statistics from the largest available brain and blood cis-eQTL meta-analyses applied to the largest clinically-adjudicated AD GWAS.MethodsWe implemented the OTTERS TWAS pipeline to predict gene expression using the largest available cis-eQTL data from cortical brain tissue (MetaBrain; N = 2683) and blood (eQTLGen; N = 31,684), and then applied these models to AD-GWAS data (Cases = 21,982; Controls = 44,944).ResultsWe identified and validated five novel gene associations in cortical brain tissue (PRKAG1, C3orf62, LYSMD4, ZNF439, SLC11A2) and six genes proximal to known AD-related GWAS loci (Blood: MYBPC3; Brain: MTCH2, CYB561, MADD, PSMA5, ANXA11). Further, using causal eQTL fine-mapping, we generated sparse models that retained the strength of the AD-TWAS association for MTCH2, MADD, ZNF439, CYB561, and MYBPC3.ConclusionsOur comprehensive AD-TWAS discovered new gene associations and provided insights into the functional relevance of previously associated variants, which enables us to further understand the genetic architecture underlying AD risk.

Unsupervised Ensemble Learning for Efficient Integration of Pre-trained Polygenic Risk Scores

The growing availability of pre-trained polygenic risk score (PRS) models has enabled their integration into real-world applications, reducing the need for extensive data labeling, training, and calibration. However, selecting the most suitable PRS model for a specific target population remains challenging, due to issues such as limited transferability, data heterogeneity, and the scarcity of observed phenotype in real-world settings. Ensemble learning offers a promising avenue to enhance the predictive accuracy of genetic risk assessments, but most existing methods often rely on observed phenotype data or additional genome-wide association studies (GWAS) from the target population to optimize ensemble weights, limiting their utility in real-time implementation. Here, we present the UN supervised en Semble PRS ( UNSemblePRS ), an unsupervised ensemble learning framework, that combines pre-trained PRS models without requiring phenotype data or summaries from the target population. Unlike traditional supervised approaches, UNSemblePRS aggregates models based on prediction concordance across a curated subset of candidate PRS models. We evaluated UNSemblePRS using both continuous and binary traits in the All of Us database, demonstrating its scalability and robust performance across diverse populations. These results underscore UNSemblePRS as an accessible tool for integrating PRS models into real-world contexts, offering broad applicability as the availability of PRS models continues to expand.

Genetic predictions of eye and hair colour in the Danish population

Genetic predictions of eye and hair colour are prominent examples of forensic DNA phenotyping that can help resolve criminal cases. The advent of high-throughput genotyping technologies in forensic genetics opens up the possibility of applying polygenic risk scores in forensic settings. In this work, we compare the performance of HIrisPlex with PRSice-2 in predicting eye and hair colour to gain insights into the relative benefits of new approaches. Predictions were carried out on 584 Danish high school students for which genetic and self-reported phenotype data were available. Prediction of brown eye colour was very accurate (AUC = 0.98), followed by blue eye colour (AUC = 0.82), while it failed for intermediate eye colour (AUC = 0.57). As for hair colour, red and black were overall better predicted than blond and brown, and PRSice-2 performed better in all but the black hair colour. Despite the limitations of the study, HIrisPlex exhibited its usual high performance in the prediction of brown and blue eye colour, as well as red and black hair colour. However, PRSice-2 offered overall improvements in hair colour prediction over HIrisPlex suggesting that there is room for improvement in forensic DNA phenotyping by using polygenic risk scores.

Gene-environment correlation: the role of family environment in academic development

Academic achievement is partly heritable and highly polygenic. However, genetic effects on academic achievement are not independent of environmental processes. We investigated whether aspects of the family environment mediated genetic effects on academic achievement across development. Our sample included 5151 children who participated in the Twins Early Development Study, as well as their parents and teachers. Data on academic achievement and family environments (parenting, home environments, and geocoded indices of neighbourhood characteristics) were available at ages 7, 9, 12 and 16. We computed educational attainment polygenic scores (PGS) and further separated genetic effects into cognitive and noncognitive PGS. Three core findings emerged. First, aspects of the family environment, but not the wider neighbourhood context, consistently mediated the PGS effects on achievement across development-accounting for up to 34.3% of the total effect. Family characteristics mattered beyond socio-economic status. Second, family environments were more robustly linked to noncognitive PGS effects on academic achievement than cognitive PGS effects. Third, when we investigated whether environmental mediation effects could also be observed when considering differences between siblings, adjusting for family fixed effects, we found that environmental mediation was nearly exclusively observed between families. This is consistent with the proposition that family environmental contexts contribute to academic development via passive gene-environment correlation processes or genetic nurture. Our results show how parents tend to shape environments that foster their children's academic development partly based on their own genetic disposition, particularly towards noncognitive skills, rather than responding to each child's genetic disposition.

Meta-analysis of genome-wide associations and polygenic risk prediction for atrial fibrillation in more than 180,000 cases

Atrial fibrillation (AF) is the most common heart rhythm abnormality and is a leading cause of heart failure and stroke. This large-scale meta-analysis of genome-wide association studies increased the power to detect single-nucleotide variant associations and found more than 350 AF-associated genetic loci. We identified candidate genes related to muscle contractility, cardiac muscle development and cell-cell communication at 139 loci. Furthermore, we assayed chromatin accessibility using assay for transposase-accessible chromatin with sequencing and histone H3 lysine 4 trimethylation in stem cell-derived atrial cardiomyocytes. We observed a marked increase in chromatin accessibility for our sentinel variants and prioritized genes in atrial cardiomyocytes. Finally, a polygenic risk score (PRS) based on our updated effect estimates improved AF risk prediction compared to the CHARGE-AF clinical risk score and a previously reported PRS for AF. The doubling of known risk loci will facilitate a greater understanding of the pathways underlying AF.

A multi-ancestry genome-wide association study and evaluation of polygenic scores of LDL-C levels

The genetic determinants of low-density lipoprotein cholesterol (LDL-C) levels in blood have been predominantly explored in European populations and remain poorly understood in Middle Eastern populations. We investigated the genetic architecture of LDL-C variation in Qatar by conducting a genome-wide association study (GWAS) on serum LDL-C levels using whole genome sequencing data of 13,701 individuals (discovery; n = 5,939, replication; n = 7,762) from the population-based Qatar Biobank (QBB) cohort. We replicated 168 previously reported loci from the largest LDL-C GWAS by the Global Lipids Genetics Consortium (GLGC), with high correlation in allele frequencies (R2 = 0.77) and moderate correlation in effect sizes (Beta; R2 = 0.53). We also performed a multi-ancestry meta-analysis with the GLGC study using MR-MEGA (Meta-Regression of Multi-Ethnic Genetic Association) and identified one novel LDL-C-associated locus; rs10939663 (SLC2A9; genomic control-corrected P = 1.25 × 10-8). Lastly, we developed Qatari-specific polygenic score (PGS) panels and tested their performance against PGS derived from other ancestries. The multi-ancestry-derived PGS (PGS000888) performed best at predicting LDL-C levels, whilst the Qatari-derived PGS showed comparable performance. Overall, we report a novel gene associated with LDL-C levels, which may be explored further to decipher its potential role in the etiopathogenesis of cardiovascular diseases. Our findings also highlight the importance of population-based genetics in developing PGS for clinical utilization.

From genetic disposition to academic achievement: The mediating role of non-cognitive skills across development

Genetic effects on academic achievement are likely to capture environmental, developmental, and psychological processes. How these processes contribute to translating genetic dispositions into observed academic achievement remains critically under-investigated. Here, we examined the role of non-cognitive skills-e.g., motivation, attitudes and self-regulation-in mediating education-associated genetic effects on academic achievement across development. Data were collected from 5,016 children enrolled in the Twins Early Development Study at ages 7, 9, 12, and 16, as well as their parents and teachers. We found that non-cognitive skills mediated polygenic score effects on academic achievement across development, and longitudinally, accounting for up to 64% of the total effects. Within-family analyses highlighted the contribution of non-cognitive skills beyond genetic, environmental and demographic factors that are shared between siblings, accounting for up to 83% of the total mediation effect, likely reflecting evocative/active gene-environment correlation. Our results underscore the role of non-cognitive skills in academic development in how children evoke and select experiences that align with their genetic propensity.

PennPRS: a centralized cloud computing platform for efficient polygenic risk score training in precision medicine

Polygenic risk scores (PRS) are becoming increasingly vital for risk prediction and stratification in precision medicine. However, PRS model training presents significant challenges for broader adoption of PRS, including limited access to computational resources, difficulties in implementing advanced PRS methods, and availability and privacy concerns over individual-level genetic data. Cloud computing provides a promising solution with centralized computing and data resources. Here we introduce PennPRS (https://pennprs.org), a scalable cloud computing platform for online PRS model training in precision medicine. We developed novel pseudo-training algorithms for multiple PRS methods and ensemble approaches, enabling model training without requiring individual-level data. These methods were rigorously validated through extensive simulations and large-scale real data analyses involving over 6,000 phenotypes across various data sources. PennPRS supports online single- and multi-ancestry PRS training with seven methods, allowing users to upload their own data or query from more than 27,000 datasets in the GWAS Catalog, submit jobs, and download trained PRS models. Additionally, we applied our pseudo-training pipeline to train PRS models for over 8,000 phenotypes and made their PRS weights publicly accessible. In summary, PennPRS provides a novel cloud computing solution to improve the accessibility of PRS applications and reduce disparities in computational resources for the global PRS research community.

Blood-Based EWAS of Asthma Polygenic Burden in The Netherlands Twin Register

Asthma, a chronic respiratory condition characterized by airway inflammation, affects millions of individuals worldwide. Challenges remain in asthma prediction and diagnosis from its complex etiology involving genetic and environmental factors. Here, we investigated the relationship between genome-wide DNA methylation and genetic risk for asthma quantified via polygenic scores in two cohorts from the Netherlands Twin Register; one enriched with asthmatic families measured on the Illumina EPIC array (n = 526) and a general population cohort measured on the Illumina HM450K array (n = 2680). We performed epigenome-wide association studies of asthma polygenic scores in each cohort with results combined through meta-analysis (total samples = 3206). The EWAS meta-analysis identified 63 significantly associated CpGs, (following Bonferroni correction, α = 0.05/358,316). An investigation of previous mQTL associations identified 48 mQTL associations between 24 unique CpGs and 48 SNPs, of which two SNPs have previous associations with asthma. Enrichment analysis using the 63 significant CpGs highlighted previous associations with ancestry, smoking, and air pollution. A dizygotic twin within-pair analysis of the 63 CpGs revealed similar directional effects between the two cohorts in 33 of the 63 CpGs. These findings further characterize the intricate relationship between DNA methylation and genetics relative to asthma.

PNL: a software to build polygenic risk scores using a super learner approach based on PairNet, a Convolutional Neural Network

SUMMARY

Polygenic risk scores (PRSs) hold promise for early disease diagnosis and personalized treatment, but their overall discriminative power remains limited for many diseases in the general population. As a result, numerous novel PRS modeling techniques have been developed to improve predictive performance, but determining the most effective method for a specific application remains uncertain until tested. Hence, we introduce a novel, versatile tool for building an optimized PRS model by integrating candidate models from multiple existing PRS building methods that use target population data and/or incorporating information from other populations through a trans-ethnic approach. Our tool, PNL is based on PairNet algorithm, a Convolutional Neural Network with low computation complexity through simple paring operation. In the case studies for asthma, type 2 diabetes, and vertigo, the optimal PRS model generated with PNL using only Taiwan biobank (TWB) data achieved Area Under the Curves (AUCs) that matched or improved the best results using other methods individually. Incorporating the UK Biobank data (UKBB) data further improved performance of PNL for asthma and type 2 diabetes. For vertigo, unlike the other diseases, individual method analysis showed that UKBB data alone generally produced lower AUCs compared to TWB data alone. As a result, incorporating UKBB data did not improve AUC with PNL, suggesting that increasing the number of candidate models does not necessarily result in higher AUC values, alleviating concerns about overfitting.

AVAILABILITY AND IMPLEMENTATION

The python code for PairNet algorithm incorporated in PNL is freely available on: https://github.com/FannLab/pairnet. An archived, citable version is stored on: https://doi.org/10.5281/zenodo.14838227.

Polygenic modelling and machine learning approaches in pharmacogenomics: Importance in downstream analysis of genome-wide association study data

Genome-wide association studies (GWAS) have identified genetic variations associated with adverse drug effects in pharmacogenomics (PGx) research. However, interpreting the biological implications of these associations remains a challenge. This review highlights 2 promising post-GWAS methods for PGx. First, we discuss the polygenic architecture of the PGx traits, especially for drug-induced liver injury. Experimental modelling using multiple donors' human primary hepatocytes and human liver organoids demonstrated the polygenic architecture of drug-induced liver injury susceptibility and found biological vulnerability in genetically high-risk tissue donors. Second, we discuss the challenges of interpreting the roles of variants in noncoding regions. Beyond methods involving expression quantitative trait locus analysis and massively parallel reporter assays, we suggest the use of in silico mutagenesis through machine learning methods to understand the roles of variants in transcriptional regulation. This review underscores the importance of these post-GWAS methods in providing critical insights into PGx, potentially facilitating drug development and personalized treatment.

Multi-polygenic score prediction of mathematics, reading, and language abilities independent of general cognitive ability

Specific cognitive abilities (SCA) correlate genetically about 0.50, which underpins general cognitive ability (g), but it also means that there is considerable genetic specificity. If g is not controlled, then genomic prediction of specific cognitive abilities is not truly specific because they are all perfused with g. Here, we investigated the heritability of mathematics, reading, and language ability independent of g (SCA.g) using twins and DNA, and the extent to which multiple genome-wide polygenic scores (multi-PGS) can jointly predict these SCA.g as compared to SCA uncorrected for g. We created SCA and SCA.g composites from a battery of 14 cognitive tests administered at age 12 to 5,000 twin pairs in the Twins Early Development Study (TEDS). Univariate twin analyses yielded an average heritability estimate of 40% for SCA.g, compared to 53% for uncorrected SCA. Using genome-wide SNP genotypes, average SNP-based heritabilities were 26% for SCA.g and 35% for SCA. We then created multi-PGS from at least 50 PGS to predict each SCA and SCA.g using elastic net penalised regression models. Multi-PGS predicted 4.4% of the variance of SCA.g on average, compared to 11.1% for SCA uncorrected for g. The twin, SNP and PGS heritability estimates for SCA.g provide further evidence that the heritabilities of SCA are not merely a reflection of g. Although the relative reduction in heritability from SCA to SCA.g was greater for PGS heritability than for twin or SNP heritability, this decrease is likely due to the paucity of PGS for SCA. We hope that these results encourage researchers to conduct genome-wide association studies of SCA, and especially SCA.g, that can be used to predict PGS profiles of SCA strengths and weaknesses independent of g.

Developing and validating a comprehensive polygenic risk score to enhance keratoconus risk prediction

PURPOSE

This study aimed to develop and validate a comprehensive polygenic risk score (PRS) for keratoconus, enhancing the predictive accuracy for identifying individuals at increased risk, which is crucial for preventing keratoconus-associated visual impairment such as post-Laser-assisted in situ keratomileusis (LASIK) ectasia.

METHODS

We applied a multi-trait analysis approach (MTAG) to genome-wide association study data on keratoconus and quantitative keratoconus-related traits and used this to construct PRS models for keratoconus risk using several PRS methodologies. We evaluated the predictive performance of the PRSs in two biobanks: Estonian Biobank (EstBB; 375 keratoconus cases and 17 902 controls) and UK Biobank (UKB: 34 keratoconus cases and 1000 controls). Scores were compared using the area under the curve (AUC) and odds ratios (ORs) for various PRS models.

RESULTS

The PRS models demonstrated significant predictive capabilities in EstBB, with the SBayesRC model achieving the highest OR of 2.28 per standard deviation increase in PRS, with a model containing age, sex and PRS showing good predictive accuracy (AUC = 0.72). In UKB, we found that adding the best-performing PRS to a model containing corneal measurements increased the AUC from 0.84 to 0.88 (P = 0.012 for difference), with an OR of 4.26 per standard deviation increase in the PRS. These models showed improved predictive capability compared to previous keratoconus PRS.

CONCLUSION

The PRS models enhanced prediction of keratoconus risk, even with corneal measurements, showing potential for clinical use to identify individuals at high risk of keratoconus, and potentially help reduce the risk of post-LASIK ectasia.

Modeling gene interactions in polygenic prediction via geometric deep learning

Polygenic risk score (PRS) is a widely used approach for predicting individuals' genetic risk of complex diseases, playing a pivotal role in advancing precision medicine. Traditional PRS methods, predominantly following a linear structure, often fall short in capturing the intricate relationships between genotype and phenotype. In this study, we present PRS-Net, an interpretable geometric deep learning-based framework that effectively models the nonlinearity of biological systems for enhanced disease prediction and biological discovery. PRS-Net begins by deconvoluting the genome-wide PRS at the single-gene resolution and then explicitly encapsulates gene-gene interactions leveraging a graph neural network (GNN) for genetic risk prediction, enabling a systematic characterization of molecular interplay underpinning diseases. An attentive readout module is introduced to facilitate model interpretation. Extensive tests across multiple complex traits and diseases demonstrate the superior prediction performance of PRS-Net compared with a wide range of conventional PRS methods. The interpretability of PRS-Net further enhances the identification of disease-relevant genes and gene programs. PRS-Net provides a potent tool for concurrent genetic risk prediction and biological discovery for complex diseases.

Leveraging haplotype information in heritability estimation and polygenic prediction

Polygenic prediction has yet to make a major clinical breakthrough in precision medicine and psychiatry, where the application of polygenic risk scores is expected to improve clinical decision-making. Most widely used approaches for estimating polygenic risk scores are based on summary statistics from external large-scale genome-wide association studies, which rely on assumptions of matching data distributions. This may hinder the impact of polygenic risk scores in modern diverse populations due to small differences in genetic architectures. Reference-free estimators of polygenic scores are instead based on genomic best linear unbiased predictions and model the population of interest directly. We introduce a framework, named hapla, with a novel algorithm for clustering haplotypes in phased genotype data to estimate heritability and perform reference-free polygenic prediction in complex traits. We utilize inferred haplotype clusters to compute accurate heritability estimates and polygenic scores in a simulation study and the iPSYCH2012 case-cohort for depression disorders and schizophrenia. We demonstrate that our haplotype-based approach robustly outperforms standard genotype-based approaches, which can help pave the way for polygenic risk scores in the future of precision medicine and psychiatry.

Nature or nurture: genetic and environmental predictors of adiposity gain in adults

BACKGROUND

Previous prediction models for adiposity gain have not yet achieved sufficient predictive ability for clinical relevance. We investigated whether traditional and genetic factors accurately predict adiposity gain.

METHODS

A 5-year gain of ≥5% in body mass index (BMI) and waist-to-hip ratio (WHR) from baseline were predicted in mid-late adulthood individuals (median of 55 years old at baseline). Proportional hazards models were fitted in 245,699 participants from the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort to identify robust environmental predictors. Polygenic risk scores (PRS) of 5 proxies of adiposity [BMI, WHR, and three body shape phenotypes (PCs)] were computed using genetic weights from an independent cohort (UK Biobank). Environmental and genetic models were validated in 29,953 EPIC participants.

FINDINGS

Environmental models presented a remarkable predictive ability (AUCBMI: 0.69, 95% CI: 0.68-0.70; AUCWHR: 0.75, 95% CI: 0.74-0.77). The genetic geographic distribution for WHR and PC1 (overall adiposity) showed higher predisposition in North than South Europe. Predictive ability of PRSs was null (AUC: ∼0.52) and did not improve when combined with environmental models. However, PRSs of BMI and PC1 showed some prediction ability for BMI gain from self-reported BMI at 20 years old to baseline observation (early adulthood) (AUC: 0.60-0.62).

INTERPRETATION

Our study indicates that environmental models to discriminate European individuals at higher risk of adiposity gain can be integrated in standard prevention protocols. PRSs may play a robust role in predicting adiposity gain at early rather than mid-late adulthood suggesting a more important role of genetic factors in this life period.

FUNDING

French National Cancer Institute (INCA_N°2019-176) 1220, German Research Foundation (BA 5459/2-1), Instituto de Salud Carlos III (Miguel Servet Program CP21/00058).

Healthy lifestyle practice correlates with decreased obesity prevalence in individuals with high polygenic risk: TMM CommCohort study

Obesity and overweight, fundamental components of the metabolic syndrome, predispose individuals to lifestyle-related diseases. The extent to which adopting healthy lifestyles can reduce obesity risk, even in those with a high genetic risk, remains uncertain. Our aim was to assess the extent to which lifestyle modifications can improve outcomes in individuals with a high polygenic score (PGS) for obesity. We quantified the genetic risk of obesity using PGSs. Four datasets from the Tohoku Medical Megabank Community-Based Cohort (TMM CommCohort) were employed in the study. One dataset (n = 9958) was used to select the best model for calculating PGS. The remaining datasets (total n = 69,341) were used in a meta-analysis to validate the model and to evaluate associated risks. The odds ratio (OR) for obesity risk in the intermediate (11th-90th percentiles in the dataset) and high PGS categories (91st-100th) was 2.27 [95% confidence intervals: 2.12-2.44] and 4.83 [4.45-5.25], respectively, compared to that in the low PGS category (1st-10th). Trend analysis showed that an increase in leisure-time physical activity was significantly associated with reduced obesity risk across all genetic risk categories, representing an OR of 0.9 [0.87-0.94] even among individuals in the high PGS category. Similarly, sodium intake displayed a positive association with obesity across all genetic risk categories, yielding an OR of 1.24 [1.17-1.31] in the high PGS category. The risk of obesity was linked to the adoption of healthy lifestyles, even in individuals with high PGS. Our results may provide perspectives for integrating PGSs into preventive medicine.

Unidirectional and bidirectional causation between smoking and blood DNA methylation: evidence from twin-based Mendelian randomisation

Cigarette smoking is associated with numerous differentially-methylated genomic loci in multiple human tissues. These associations are often assumed to reflect the causal effects of smoking on DNA methylation (DNAm), which may underpin some of the adverse health sequelae of smoking. However, prior causal analyses with Mendelian Randomisation (MR) have found limited support for such effects. Here, we apply an integrated approach combining MR with twin causal models to examine causality between smoking and blood DNAm in the Netherlands Twin Register (N = 2577). Analyses revealed potential causal effects of current smoking on DNAm at > 500 sites in/near genes enriched for functional pathways relevant to known biological effects of smoking (e.g., hemopoiesis, cell- and neuro-development, and immune regulation). Notably, we also found evidence of reverse and bidirectional causation at several DNAm sites, suggesting that variation in DNAm at these sites may influence smoking liability. Seventeen of the loci with putative effects of DNAm on smoking showed highly specific enrichment for gene-regulatory functional elements in the brain, while the top three sites annotated to genes involved in G protein-coupled receptor signalling and innate immune response. These novel findings are partly attributable to the analyses of current smoking in twin models, rather than lifetime smoking typically examined in MR studies, as well as the increased statistical power achieved using multiallelic/polygenic scores as instrumental variables while controlling for potential horizontal pleiotropy. This study highlights the value of twin studies with genotypic and DNAm data for investigating causal relationships of DNAm with health and disease.

Exploring the genetic architecture of brain structure and ADHD using polygenic neuroimaging-derived scores

Genome-wide association studies (GWAS) have provided valuable insights into the genetic basis of neuropsychiatric disorders and highlighted their complexity. Careful consideration of the polygenicity and complex genetic architecture could aid in the understanding of the underlying brain mechanisms. We introduce an innovative approach to polygenic scoring, utilizing imaging-derived phenotypes (IDPs) to predict a clinical phenotype. We leveraged IDP GWAS data from the UK Biobank, to create polygenic imaging-derived scores (PIDSs). As a proof-of-concept, we assessed genetic variations in brain structure between individuals with ADHD and unaffected controls across three NeuroIMAGE waves (n = 954). Out of the 94 PIDS, 72 exhibited significant associations with their corresponding IDPs in an independent sample. Notably, several global measures, including cerebellum white matter, cerebellum cortex, and cerebral white matter, displayed substantial variance explained for their respective IDPs, ranging from 3% to 5.7%. Conversely, the associations between each IDP and the clinical ADHD phenotype were relatively weak. These findings highlight the growing power of GWAS in structural neuroimaging traits, enabling the construction of polygenic scores that accurately reflect the underlying polygenic architecture. However, to establish robust connections between PIDS and behavioral or clinical traits such as ADHD, larger samples are needed. Our novel approach to polygenic risk scoring offers a valuable tool for researchers in the field of psychiatric genetics.

Healthy First-Degree Relatives From Multiplex Families vs Simplex Families Have Higher Subclinical Intestinal Inflammation, a Distinct Fecal Microbial Signature, and Harbor a Higher Risk of Developing Crohn's Disease

BACKGROUND & AIMS

Unaffected first-degree relatives (FDRs) from families with ≥2 affected FDRs with Crohn's disease (CD, multiplex families) have a high risk of developing CD, although the underlying mechanisms driving this risk are poorly understood. We aimed to identify differences in biomarkers between FDRs from multiplex vs simplex families and investigate the risk of future CD onset accounting for potential confounders.

METHODS

We assessed the Crohn's and Colitis Canada Genetic Environmental Microbial cohort of healthy FDRs of patients with CD. Genome-wide CD-polygenic risk scores, urinary fractional excretion of lactulose-to-mannitol ratio, fecal calprotectin (FCP), and fecal 16S ribosomal RNA microbiome were measured at recruitment. Associations between CD multiplex status and baseline biomarkers were determined using generalized estimating equations models. Cox models were used to assess the risk of future CD onset.

RESULTS

There were 4051 participants from simplex families and 334 from CD multiplex families. CD multiplex status was significantly associated with higher baseline FCP (P = .026) but not with baseline CD-polygenic risk scores or the lactulose-to-mannitol ratio. Three bacterial genera were found to be differentially abundant between both groups. CD multiplex status at recruitment was independently associated with an increased risk of developing CD (adjusted hazard ratio, 3.65; 95% confidence interval, 2.18-6.11, P < .001).

CONCLUSION

Within FDRs of patients with CD, participants from multiplex families had a 3-fold increased risk of CD onset, a higher FCP, and an altered bacterial composition, but not genetic burden or altered gut permeability. These results suggest that putative environmental factors might be enriched in FDRs from multiplex families.

Predicting Lung Cancer in Korean Never-Smokers With Polygenic Risk Scores

In the last few decades, genome-wide association studies (GWAS) with more than 10,000 subjects have identified several loci associated with lung cancer and these loci have been used to develop novel risk prediction tools for cancer. The present study aimed to establish a lung cancer prediction model for Korean never-smokers using polygenic risk scores (PRSs); PRSs were calculated using a pruning-thresholding-based approach based on 11 genome-wide significant single nucleotide polymorphisms (SNPs). Overall, the odds ratios tended to increase as PRSs were larger, with the odds ratio of the top 5% PRSs being 1.71 (95% confidence interval: 1.31-2.23) using the 40%-60% percentile group as the reference, and the area under the curve (AUC) of the prediction model being of 0.76 (95% confidence interval: 0.747-0.774). The receiver operating characteristic (ROC) curves of the prediction model with and without PRSs as covariates were compared using DeLong's test, and a significant difference was observed. Our results suggest that PRSs can be valuable tools for predicting the risk of lung cancer.

Polygenic scores and their applications in kidney disease

Genome-wide association studies (GWAS) have uncovered thousands of risk variants that individually have small effects on the risk of human diseases, including chronic kidney disease, type 2 diabetes, heart diseases and inflammatory disorders, but cumulatively explain a substantial fraction of disease risk, underscoring the complexity and pervasive polygenicity of common disorders. This complexity poses unique challenges to the clinical translation of GWAS findings. Polygenic scores combine small effects of individual GWAS risk variants across the genome to improve personalized risk prediction. Several polygenic scores have now been developed that exhibit sufficiently large effects to be considered clinically actionable. However, their clinical use is limited by their partial transferability across ancestries and a lack of validated models that combine polygenic, monogenic, family history and clinical risk factors. Moreover, prospective studies are still needed to demonstrate the clinical utility and cost-effectiveness of polygenic scores in clinical practice. Here, we discuss evolving methods for developing polygenic scores, best practices for validating and reporting their performance, and the study designs that will empower their clinical implementation. We specifically focus on the polygenic scores relevant to nephrology and other chronic, complex diseases and review their key limitations, necessary refinements and potential clinical applications.

Genome-wide meta-analyses of non-response to antidepressants identify novel loci and potential drugs

Antidepressants exhibit a considerable variation in efficacy, and increasing evidence suggests that individual genetics contribute to antidepressant treatment response. Here, we combined data on antidepressant non-response measured using rating scales for depressive symptoms, questionnaires of treatment effect, and data from electronic health records, to increase statistical power to detect genomic loci associated with non-response to antidepressants in a total sample of 135,471 individuals prescribed antidepressants (25,255 non-responders and 110,216 responders). We performed genome-wide association meta-analyses, genetic correlation analyses, leave-one-out polygenic prediction, and bioinformatics analyses for genetically informed drug prioritization. We identified two novel loci (rs1106260 and rs60847828) associated with non-response to antidepressants and showed significant polygenic prediction in independent samples. Genetic correlation analyses show positive associations between non-response to antidepressants and most psychiatric traits, and negative associations with cognitive traits and subjective well-being. In addition, we investigated drugs that target proteins likely involved in mechanisms underlying antidepressant non-response, and shortlisted drugs that warrant further replication and validation of their potential to reduce depressive symptoms in individuals who do not respond to first-line antidepressant medications. These results suggest that meta-analyses of GWAS utilizing real-world measures of treatment outcomes can increase sample sizes to improve the discovery of variants associated with non-response to antidepressants.

One score to rule them all: regularized ensemble polygenic risk prediction with GWAS summary statistics

Ensemble learning has been increasingly popular for boosting the predictive power of polygenic risk scores (PRS), with almost every recent multi-ancestry PRS approach employing ensemble learning as a final step. Existing ensemble approaches rely on individual-level data for model training, which severely limits their real-world applications, especially in non-European populations without sufficient genomic samples. Here, we introduce a statistical framework to construct regularized ensemble PRS, which allows us to combine a large number of candidate PRS models using only summary statistics from genome-wide association studies. We demonstrate its robust and substantial improvement over many existing PRS models in both within- and cross-ancestry applications. We believe this is truly "one score to rule them all" due to its capability to continuously combine newly developed PRS models with existing models to improve prediction performance, which makes it a universal approach that should always be employed in future PRS applications.

Mapping the relative accuracy of cross-ancestry prediction

The overwhelming majority of participants in genome-wide association studies (GWAS) have European (EUR) ancestry, and polygenic scores (PGS) derived from EURs often perform poorly in non-EURs. Previous studies suggest that between-ancestry differences in allele frequencies and linkage disequilibrium are significant contributors to the poor portability of PGS in cross-ancestry prediction. We hypothesize that the portability of (local) PGS varies significantly over the genome. Therefore, we develop a method, MC-ANOVA, to estimate the loss of accuracy in cross-ancestry prediction attributable to allele frequency and linkage disequilibrium differences between ancestries. Using data from the UK Biobank we develop PGS relative accuracy (RA) maps quantifying the local portability of EUR-derived PGS in non-EUR ancestries. We report substantial variability in RA along the genome, suggesting that even in ancestries with low overall RA of EUR-derived effects (e.g., African), there are regions with high RA. We substantiate our findings using six complex traits, which show that EUR-derived effects from regions where MC-ANOVA predicts high RA also have high empirical RA in real PGS. We provide software implementing MC-ANOVA and RA maps for several non-EUR ancestries. These maps can be used to interpret similarities and differences in GWAS results between groups and to improve cross-ancestry prediction.

Examining the role of common variants in rare neurodevelopmental conditions

Although rare neurodevelopmental conditions have a large Mendelian component1, common genetic variants also contribute to risk2,3. However, little is known about how this polygenic risk is distributed among patients with these conditions and their parents nor its interplay with rare variants. It is also unclear whether polygenic background affects risk directly through alleles transmitted from parents to children, or whether indirect genetic effects mediated through the family environment4 also play a role. Here we addressed these questions using genetic data from 11,573 patients with rare neurodevelopmental conditions, 9,128 of their parents and 26,869 controls. Common variants explained around 10% of variance in risk. Patients with a monogenic diagnosis had significantly less polygenic risk than those without, supporting a liability threshold model5. A polygenic score for neurodevelopmental conditions showed only a direct genetic effect. By contrast, polygenic scores for educational attainment and cognitive performance showed no direct genetic effect, but the non-transmitted alleles in the parents were correlated with the child's risk, potentially due to indirect genetic effects and/or parental assortment for these traits4. Indeed, as expected under parental assortment, we show that common variant predisposition for neurodevelopmental conditions is correlated with the rare variant component of risk. These findings indicate that future studies should investigate the possible role and nature of indirect genetic effects on rare neurodevelopmental conditions, and consider the contribution of common and rare variants simultaneously when studying cognition-related phenotypes.

Advancements and limitations in polygenic risk score methods for genomic prediction: a scoping review

This scoping review aims to identify and evaluate the landscape of Polygenic Risk Score (PRS)-based methods for genomic prediction from 2013 to 2023, highlighting their advancements, key concepts, and existing gaps in knowledge, research, and technology. Over the past decade, various PRS-based methods have emerged, each employing different statistical frameworks aimed at enhancing prediction accuracy, processing speed and memory efficiency. Despite notable advancements, challenges persist, including unrealistic assumptions regarding sample sizes and the polygenicity of traits necessary for accurate predictions, as well as limitations in exploring hyper-parameter spaces and considering environmental interactions. We included studies focusing on PRS-based methods for risk prediction that underwent methodological evaluations using valid approaches and released computational tools/software. Additionally, we restricted our selection to studies involving human participants that were published in English language. This review followed the standard protocol recommended by Joanna Briggs Institute Reviewer's Manual, systematically searching Ovid MEDLINE, Ovid Embase, Scopus and Web of Science databases. Additionally, searches included grey literature sources like pre-print servers such as bioRxiv, and articles recommended by experts to ensure comprehensive and diverse coverage of relevant records. This study identified 34 studies detailing 37 genomic prediction methods, the majority of which rely on linkage disequilibrium (LD) information and necessitate hyper-parameter tuning. Nine methods integrate functional/gene annotation, while 12 are suitable for cross-ancestry genomic prediction, with only one considering gene-environment (GxE) interaction. While some methods require individual-level data, most leverage summary statistics, offering flexibility. Despite progress, challenges remain. These include computational complexity and the need for large sample sizes for high prediction accuracy. Furthermore, recent methods exhibit varying effectiveness across traits, with absolute accuracies often falling short of clinical utility. Transferability across ancestries varies, influenced by trait heritability and diversity of training data, while handling admixed populations remains challenging. Additionally, the absence of standard error measurements for individual PRSs, crucial in clinical settings, underscores a critical gap. Another issue is the lack of customizable graphical visualization tools among current software packages. While genomic prediction methods have advanced significantly, there is still room for improvement. Addressing current challenges and embracing future research directions will lead to the development of more universally applicable, robust, and clinically relevant genomic prediction tools.

Influence of Genetic Ancestry on Gene-Environment Interactions of Polygenic Risk and Sociocultural Factors: Results from the Hispanic Community Health Study/Study of Latinos

Background

Many present analyses of Hispanic/Latino populations in epidemiologic research aggregate all members of this ethnic group, despite immense diversity in genetic backgrounds, environment, and culture between and across Hispanic/Latino background groups. Using the Hispanic Community Health Study/Study of Latinos (HCHS/SOL), we examined the role of self-identified background group and genetic ancestry proportions in gene-environment interactions influencing the relationship between body mass index (BMI) and a polygenic score for BMI (PGSBMI).

Methods

Weighted univariate and multivariable generalized linear models were executed to compare the effects of environmental variables identified a priori by McArdle et al. 2021. Both Amerindigenous (AME) ancestry proportion and background group identity were statistically modeled as confounders both through stratified and joint analyses to understand their influence on the relationship between BMI and PGSBMI, while incorporating gene-environment interactions of PGS x diet and PGS x age-at-immigration.

Results

After complex survey weighting, 7,075 participants remained in the analytic sample, representing individuals of six background groups: Central American, Cuban, Dominican, Mexican, Puerto Rican, and South American. The distributions of key environmental and sociocultural variables were heterogeneous between Hispanic/Latino background groups. Associations of these variables with AME ancestry were similarly heterogeneous upon stratification, indicating confounding by background group. In a predictive model for BMI incorporating health, immigration, and environmental variables, PGSBMI performance decreased with increasing AME ancestry proportion. In this model, most statistically significant GxE interactions lost significance after ancestry and background stratification, except for PGS x age-at-immigration interactions in some strata: Mexican background individuals born in the US compared to those >=21 years old at migration (β=1.33, p<0.01), Dominican background individuals 6-12 years old at migration compared to those >=21 years old at migration (β=4.38, p<0.001), and Cuban background individuals 0-5 years old at migration compared to those >=21 years old at migration (β=2.20, p=0.015), where US-born includes individuals born in the US 50 states/DC.

Conclusions

Controlling for self-identified background group identity and genetic ancestry did not eliminate statistically significant differences in interactions between AME ancestry and environmental variables in certain strata of AME ancestry among some Hispanic/Latino background groups in HCHS/SOL.

Unidirectional and Bidirectional Causation between Smoking and Blood DNA Methylation: Evidence from Twin-based Mendelian Randomisation

Cigarette smoking is associated with numerous differentially-methylated genomic loci in multiple human tissues. These associations are often assumed to reflect the causal effects of smoking on DNA methylation (DNAm), which may underpin some of the adverse health sequelae of smoking. However, prior causal analyses with Mendelian Randomisation (MR) have found limited support for such effects. Here, we apply an integrated approach combining MR with twin causal models to examine causality between smoking and blood DNAm in the Netherlands Twin Register (N=2577). Analyses revealed potential causal effects of current smoking on DNAm at >500 sites in/near genes enriched for functional pathways relevant to known biological effects of smoking (e.g., hemopoiesis, cell- and neuro-development, and immune regulation). Notably, we also found evidence of reverse and bidirectional causation at several DNAm sites, suggesting that variation in DNAm at these sites may influence smoking liability. Seventeen of the loci with putative effects of DNAm on smoking showed highly specific enrichment for gene-regulatory functional elements in the brain, while the top three sites annotated to genes involved in G protein-coupled receptor signalling and innate immune response. These novel findings are partly attributable to the analyses of current smoking in twin models, rather than lifetime smoking typically examined in MR studies, as well as the increased statistical power achieved using multiallelic/polygenic scores as instrumental variables while controlling for potential horizontal pleiotropy. This study highlights the value of twin studies with genotypic and DNAm data for investigating causal relationships of DNAm with health and disease.

Assessing the predictive efficacy of European-based systolic blood pressure polygenic risk scores in diverse Brazilian cohorts

Despite the identification of numerous genetic variants affecting SBP in European populations, their applicability in admixed populations remains unclear. This study evaluates the predictive efficacy of a systolic blood pressure (SBP) polygenic risk score (PRS), derived from the UK Biobank data, in two Brazilian cohorts. We analyzed 944 K genetic variants consistent across an independent UK Biobank dataset, Brazilian cohorts, and HapMap database. Results show a significant association between increased PRS and SBP, as well as hypertension, in each study groups analyzed. An increase of one standard deviation in the PRS showed a significant association with SBP (β [95% CI] (mmHg) = 5.2 [5.1-5.3], 2.8 [2.1-3.5] and 2.6 [2.2-3.0]) and hypertension (odds ratio (OR) [95% CI] = 1.56 [1.54-1.56], 1.28 [1.2-1.4] and 1.47 [1.3-1.6]) in an independent UKB dataset, Baependi, and Pelotas, respectively. The associations were weaker in the Brazilian samples and the reduced association was noticeable in the Pelotas vs. the UK comparison for hypertension stages 1 and 2 (OR [95% CI] = 2.1 [1.5-3.1] and 3.0 [1.9-4.7] vs. 2.5 [2.2-2.8] and 4.9 [4.4-5.6]), whereas the Baependi data showed no significance for stage 1 hypertension. This trend mirrors findings in homogeneous African and Asian populations with diverse genetic architecture, highlighting the limitations of European-based PRS also in admixed populations. These insights are crucial for developing tailored disease prevention and management strategies in ethnically diverse groups.

Sequence variants associated with BMI affect disease risk through BMI itself

Mendelian Randomization studies indicate that BMI contributes to various diseases, but it's unclear if this is entirely mediated by BMI itself. This study examines whether disease risk from BMI-associated sequence variants is mediated through BMI or other mechanisms, using data from Iceland and the UK Biobank. The associations of BMI genetic risk score with diseases like fatty liver disease, knee replacement, and glucose intolerance were fully attenuated when conditioned on BMI, and largely for type 2 diabetes, heart failure, myocardial infarction, atrial fibrillation, and hip replacement. Similar attenuation was observed for chronic kidney disease and stroke, though results varied. Findings were consistent across sexes, except for myocardial infarction. Residual effects may result from temporal BMI changes, pleiotropy, measurement error, non-linear relationships, non-collapsibility, or confounding. The attenuation extent of BMI genetic risk score on disease associations suggests the potential impact of reducing BMI on disease risk.

Polygenic Risk Scores and Extreme Coronary Artery Calcium Phenotypes (CAC=0 and CAC≥1000) in Adults ≥75 Years Old: The ARIC Study

BACKGROUND

Coronary artery calcium (CAC) is heterogeneous in older age and is incompletely explained by traditional atherosclerotic cardiovascular disease risk factors. The extremes of subclinical atherosclerosis burden are strongly associated with either a low or high 10-year risk of incident atherosclerotic cardiovascular disease, respectively. However, the genetic underpinnings of differences in arterial aging remain unclear. We sought to determine the independent association of 2 polygenic scores for coronary heart disease (CHD) with CAC in adults ≥75 years of age.

METHODS

There were 1865 ARIC (Atherosclerosis Risk in Communities) participants who underwent genetic testing at visit 1 (1987-1989) and CAC scans at visit 7 (2018-2019). In the primary analysis, an externally derived multi-ancestry polygenic CHD risk score was calculated for both White and Black participants. Results were confirmed using a separate ARIC-derived polygenic CHD risk score, including ≥6 million variants computed for White participants. We used multivariable logistic regression models to assess the association of polygenic CHD risk with CAC, after adjusting for baseline, time-averaged lifestyle, traditional risk factors, and local ancestry principal components.

RESULTS

In the primary analysis, the average age was 80.6 years old, 61.6% were women, and the median CAC score was 246 (189 participants with CAC=0, 364 participants with CAC≥1000). Compared with persons below the 20th percentile of polygenic CHD risk, persons with polygenic-CHD risk above the 80th percentile had 82% lower odds of having CAC=0 (odds ratio, 0.18 [95% CI, 0.09-0.37]) and had >4-fold higher odds of CAC≥1000 (odds ratio, 4.77 [95% CI, 2.88-7.88]). On a continuous scale, each SD increment increase in the polygenic risk score was associated with a 78% higher CAC score. Results were nearly identical using a second confirmatory polygenic CHD risk score in White participants.

CONCLUSIONS

Polygenic CHD risk is robustly associated with a lower prevalence of CAC=0 and a higher prevalence of CAC≥1000 in adults ≥75 years of age, beyond lifestyle and traditional risk factors. These results suggest a heritable contribution to distinct healthy and unhealthy arterial aging phenotypes that persist throughout the life course.

Identification of women at high risk of postpartum psychiatric episodes: A population-based study quantifying relative and absolute risks following exposure to selected risk factors and genetic liability

BACKGROUND

We quantified relative and absolute risks of postpartum psychiatric episodes (PPE) following risk factors: Young age, past personal or family history of psychiatric disorders, and genetic liability.

METHODS

We conducted a register-based study using the iPSYCH2012 case-cohort sample. Exposures were personal history of psychiatric episodes prior to childbirth, being a young mother (giving birth before the age of 21.5 years), having a family history of psychiatric disorders, and a high (highest quartile) polygenic score (PGS) for major depression. PPE was defined within 12 months postpartum by prescription of psychotropic medication or in- and outpatient contact to a psychiatric facility. We included primiparous women born 1981-1999, giving birth before January 1st, 2016. We conducted Cox regression to calculate hazard ratios (HRs) of PPE, absolute risks were calculated using cumulative incidence functions.

RESULTS

We included 8174 primiparous women, and the estimated baseline PPE risk was 6.9% (95% CI 6.0%-7.8%, number of PPE cases: 2169). For young mothers with a personal and family history of psychiatric disorders, the absolute risk of PPE was 21.6% (95% CI 15.9%-27.8%). Adding information on high genetic liability to depression, the risk increased to 29.2% (95% CI 21.3%-38.4%) for PPE.

CONCLUSIONS

Information on prior personal and family psychiatric episodes as well as age may assist in estimating a personalized risk of PPE. Furthermore, additional information on genetic liability could add even further to this risk assessment.

The utility of obesity polygenic risk scores from research to clinical practice: A review

Obesity represents a major public health emergency worldwide, and its etiology is shaped by a complex interplay of environmental and genetic factors. Over the last decade, polygenic risk scores (PRS) have emerged as a promising tool to quantify an individual's genetic risk of obesity. The field of PRS in obesity genetics is rapidly evolving, shedding new lights on obesity mechanisms and holding promise for contributing to personalized prevention and treatment. Challenges persist in terms of its clinical integration, including the need for further validation in large-scale prospective cohorts, ethical considerations, and implications for health disparities. In this review, we provide a comprehensive overview of PRS for studying the genetics of obesity, spanning from methodological nuances to clinical applications and challenges. We summarize the latest developments in the generation and refinement of PRS for obesity, including advances in methodologies for aggregating genome-wide association study data and improving PRS predictive accuracy, and discuss limitations that need to be overcome to fully realize its potential benefits of PRS in both medicine and public health.

Environment-by-PGS Interaction in the Classical Twin Design: An Application to Childhood Anxiety and Negative Affect

One type of genotype-environment interaction occurs when genetic effects on a phenotype are moderated by an environment; or when environmental effects on a phenotype are moderated by genes. Here we outline these types of genotype-environment interaction models, and propose a test of genotype-environment interaction based on the classical twin design, which includes observed genetic variables (polygenic scores: PGSs) that account for part of the genetic variance of the phenotype. We introduce environment-by-PGS interaction and the results of a simulation study to address statistical power and parameter recovery. Next, we apply the model to empirical data on anxiety and negative affect in children. The power to detect environment-by-PGS interaction depends on the heritability of the phenotype, and the strength of the PGS. The simulation results indicate that under realistic conditions of sample size, heritability and strength of the interaction, the environment-by-PGS model is a viable approach to detect genotype-environment interaction. In 7-year-old children, we defined two PGS based on the largest genetic association studies for 2 traits that are genetically correlated to childhood anxiety and negative affect, namely major depression (MDD) and intelligence (IQ). We find that common environmental influences on negative affect are amplified for children with a lower IQ-PGS.

Effects of polygenic risk score and sodium and potassium intake on hypertension in Asians: A nationwide prospective cohort study

Genetic factors, lifestyle, and diet have been shown to play important roles in the development of hypertension. Increased salt intake is an important risk factor for hypertension. However, research on the involvement of genetic factors in the relationship between salt intake and hypertension in Asians is lacking. We aimed to investigate the risk of hypertension in relation to sodium and potassium intake and the effects of genetic factors on their interactions. We used Korean Genome and Epidemiology Study data and calculated the polygenic risk score (PRS) for the effect of systolic and diastolic blood pressure (SBP and DBP). We also conducted multivariable logistic modeling to evaluate associations among incident hypertension, PRSSBP, PRSDBP, and sodium and potassium intake. In total, 41,351 subjects were included in the test set. The top 10% PRSSBP group was the youngest of the three groups (bottom 10%, middle, top 10%), had the highest proportion of women, and had the highest body mass index, baseline BP, red meat intake, and alcohol consumption. The multivariable logistic regression model revealed the risk of hypertension was significantly associated with higher PRSSBP, higher sodium intake, and lower potassium intake. There was significant interaction between sodium intake and PRSSBP for incident hypertension especially in sodium intake ≥2.0 g/day and PRSSBP top 10% group (OR 1.27 (1.07-1.51), P = 0.007). Among patients at a high risk of incident hypertension due to sodium intake, lifestyle modifications and sodium restriction were especially important to prevent hypertension.

Methodologies underpinning polygenic risk scores estimation: a comprehensive overview

Polygenic risk scores (PRS) have emerged as a promising tool for predicting disease risk and treatment outcomes using genomic data. Thousands of genome-wide association studies (GWAS), primarily involving populations of European ancestry, have supported the development of PRS models. However, these models have not been adequately evaluated in non-European populations, raising concerns about their clinical validity and predictive power across diverse groups. Addressing this issue requires developing novel risk prediction frameworks that leverage genetic characteristics across diverse populations, considering host-microbiome interactions and a broad range of health measures. One of the key aspects in evaluating PRS is understanding the strengths and limitations of various methods for constructing them. In this review, we analyze strengths and limitations of different methods for constructing PRS, including traditional weighted approaches and new methods such as Bayesian and Frequentist penalized regression approaches. Finally, we summarize recent advances in PRS calculation methods development, and highlight key areas for future research, including development of models robust across diverse populations by underlining the complex interplay between genetic variants across diverse ancestral backgrounds in disease risk as well as treatment response prediction. PRS hold great promise for improving disease risk prediction and personalized medicine; therefore, their implementation must be guided by careful consideration of their limitations, biases, and ethical implications to ensure that they are used in a fair, equitable, and responsible manner.