Exploring the predictive power of polygenic scores derived from genome-wide association studies: a study of 10 complex traits

Discrimination between healthy participants and people with panic disorder based on polygenic scores for psychiatric disorders and for intermediate phenotypes using machine learning

OBJECTIVE

Panic disorder is a modestly heritable condition. Currently, diagnosis is based only on clinical symptoms; identifying objective biomarkers and a more reliable diagnostic procedure is desirable. We investigated whether people with panic disorder can be reliably diagnosed utilizing combinations of multiple polygenic scores for psychiatric disorders and their intermediate phenotypes, compared with single polygenic score approaches, by applying specific machine learning techniques.

METHODS

Polygenic scores for 48 psychiatric disorders and intermediate phenotypes based on large-scale genome-wide association studies (n = 7556-1,131,881) were calculated for people with panic disorder (n = 718) and healthy controls (n = 1717). Discrimination between people with panic disorder and healthy controls was based on the 48 polygenic scores using five methods for classification: logistic regression, neural networks, quadratic discriminant analysis, random forests and a support vector machine. Differences in discrimination accuracy (area under the curve) due to an increased number of polygenic score combinations and differences in the accuracy across five classifiers were investigated.

RESULTS

All five classifiers performed relatively well for distinguishing people with panic disorder from healthy controls by increasing the number of polygenic scores. Of the 48 polygenic scores, the polygenic score for anxiety UK Biobank was the most useful for discrimination by the classifiers. In combinations of two or three polygenic scores, the polygenic score for anxiety UK Biobank was included as one of polygenic scores in all classifiers. When all 48 polygenic scores were used in combination, the greatest areas under the curve significantly differed among the five classifiers. Support vector machine and logistic regression had higher accuracy than quadratic discriminant analysis and random forests. For each classifier, the greatest area under the curve was 0.600 ± 0.030 for logistic regression (polygenic score combinations N = 14), 0.591 ± 0.039 for neural networks (N = 9), 0.603 ± 0.033 for quadratic discriminant analysis (N = 10), 0.572 ± 0.039 for random forests (N = 25) and 0.617 ± 0.041 for support vector machine (N = 11). The greatest areas under the curve at the best polygenic score combination significantly differed among the five classifiers. Random forests had the lowest accuracy among classifiers. Support vector machine had higher accuracy than neural networks.

CONCLUSIONS

These findings suggest that increasing the number of polygenic score combinations up to approximately 10 effectively improved the discrimination accuracy and that support vector machine exhibited greater accuracy among classifiers. However, the discrimination accuracy for panic disorder, when based solely on polygenic score combinations, was found to be modest.

The impact of the COVID-19 pandemic on alcohol use disorder symptoms: Testing interactions with polygenic risk

Objective: The purpose of this study was to test whether COVID impact interacts with genetic risk (polygenic risk score/PRS) to predict alcohol use disorder (AUD) symptoms. Method: Participants were n = 455 college students (79.6% female, 51% European Ancestry/EA, 24% African Ancestry/AFR, 25% Americas Ancestry/AMER) from a longitudinal study during the initial stage (March-May 2020) of the pandemic. Path models allowed for the examination of PRS and previously identified COVID-19 impact constructs. Results: There was a main effect of the AUD PRS on AUD symptoms within the EA group (β: .165, p < .01). Additionally, food/housing insecurity was predictive in the AMER group (β.295, p < .05), and greater increases in substance use were associated with AUD symptoms for EA (β:.459, p < .001) and AMER groups (β:.468, p < .001). Conclusions: Greater food/housing instability and increases in substance use, as well higher scores on PRS are associated with more AUD symptoms for some ancestral groups within this college sample.

Considerations, Caveats, and Suggestions for the Use of Polygenic Scores for Social and Behavioral Traits

Polygenic scores (PGS) are increasingly being used for prediction of social and behavioral traits, but suffer from many methodological, theoretical, and ethical concerns that profoundly limit their value. Primarily, these scores are derived from statistical correlations, carrying no inherent biological meaning, and thus may capture indirect effects. Further, the performance of these scores depends upon the diversity of the reference populations and the genomic panels from which they were derived, which consistently underrepresent minoritized populations, leading to poor fit when applied to diverse groups. There is also inherent danger of eugenic applications for the information gained from these scores, and general risk of misunderstandings that could lead to stigmatization for underrepresented groups. We urge extreme caution in use of PGS particularly for social/behavioral outcomes fraught for misinterpretation, with potential harm for the minoritized groups least likely to benefit from their use.

Genetics in psychiatry: Methods, clinical applications and future perspectives

Psychiatric disorders and related traits have a demonstrated genetic component, with heritability estimated by twin studies generally between 80% and 40%. Their pathogenesis is complex and multi-determined: environmental factors interact with a polygenic architecture, making difficult the development of models able to stratify patients or predict mental health outcomes. Despite this difficult challenge, relevant progress has been made in the field of psychiatric genetics in recent years. This review aims to present the main current methods in psychiatric genetics, their output, limitations, clinical applications, and possible future developments. Genome-wide association studies (GWASs) performed in increasingly large samples have led to the identification of replicated genetic loci associated with the risk of major psychiatric disorders, including schizophrenia and mood disorders. Statistical and biological approaches have been developed to improve our understanding of the etiopathogenetic mechanisms behind genome-wide significant associations, as well as for estimating the cumulative effect of risk variants at the individual level and the genetic overlap between different disorders, as pleiotropy is the rule rather than the exception. Clinical applications are available in the pharmacogenetics field. The main issues that remain to be addressed include improving ethnic diversity in genetic studies and the optimization of statistical power through methodological improvements, such as the definition of dimensional phenotypes with specific biological correlates and the integration of different types of omics data.

The Gender-Specific Interaction of DVL3 and GSK3β Polymorphisms on Major Depressive Disorder Susceptibility in a Chinese Han Population: A Case-Control Study

Based on the “oxidative stress hypothesis” of major depressive disorder (MDD), cells regulate their structure through the Wnt pathway. Little is known regarding the interactions of dishevelled 3 (DVL3) and glycogen synthase kinase 3 beta (GSK3β) polymorphisms with MDD. The aim of the current study was to verify the relationship between DVL3 and GSK3β genetic variants in a Chinese Han population and further to evaluate whether these interactions exhibit gender-specificity. A total of 1136 participants, consisting of 541 MDD patients and 595 healthy subjects, were recruited. Five single-nucleotide polymorphisms (SNPs) of DVL3/GSK3β were selected to assess their interaction by use of a generalized multifactor dimensionality reduction method. The genotype and haplotype frequencies of DVL3/GSK3β polymorphisms were significantly different between patients and controls for DVL3 rs1709642 ( $P<0.01$ ) and GSK3β rs334558, rs6438552, and rs2199503 ( $P<0.01$ ). In addition, our results also showed that there were significant interaction effects between DVL3 and GSK3β polymorphisms and the risk of developing MDD, particularly in women. The interaction between DVL3 (rs1709642) and GSK3β (rs334558, rs6438552) showed a cross-validation (CV) consistency of 10/10, a $P$ value of 0.001, and a testing accuracy of 59.22%, which was considered as the best generalized multifactor dimensionality reduction (GMDR) model. This study reveals the interaction between DVL3 and GSK3β polymorphisms on MDD susceptibility in a female Chinese Han population. The effect of gender should be taken into account in future studies that seek to explore the genetic predisposition to MDD relative to the DVL3 and GSK3β genes.

The Genetic Basis of Hypertriglyceridemia

PURPOSE OF REVIEW

Hypertriglyceridemia is a common dyslipidemia associated with an increased risk of cardiovascular disease and pancreatitis. Severe hypertriglyceridemia may sometimes be a monogenic condition. However, in the vast majority of patients, hypertriglyceridemia is due to the cumulative effect of multiple genetic risk variants along with lifestyle factors, medications, and disease conditions that elevate triglyceride levels. In this review, we will summarize recent progress in the understanding of the genetic basis of hypertriglyceridemia.

RECENT FINDINGS

More than 300 genetic loci have been identified for association with triglyceride levels in large genome-wide association studies. Studies combining the loci into polygenic scores have demonstrated that some hypertriglyceridemia phenotypes previously attributed to monogenic inheritance have a polygenic basis. The new genetic discoveries have opened avenues for the development of more effective triglyceride-lowering treatments and raised interest towards genetic screening and tailored treatments against hypertriglyceridemia. The discovery of multiple genetic loci associated with elevated triglyceride levels has led to improved understanding of the genetic basis of hypertriglyceridemia and opened new translational opportunities.

The Liability Threshold Model for Predicting the Risk of Cardiovascular Disease in Patients with Type 2 Diabetes: A Multi-Cohort Study of Korean Adults

Personalized risk prediction for diabetic cardiovascular disease (DCVD) is at the core of precision medicine in type 2 diabetes (T2D). We first identified three marker sets consisting of 15, 47, and 231 tagging single nucleotide polymorphisms (tSNPs) associated with DCVD using a linear mixed model in 2378 T2D patients obtained from four population-based Korean cohorts. Using the genetic variants with even modest effects on phenotypic variance, we observed improved risk stratification accuracy beyond traditional risk factors (AUC, 0.63 to 0.97). With a cutoff point of 0.21, the discrete genetic liability threshold model consisting of 231 SNPs (GLT₂₃₁) correctly classified 87.7% of 2378 T2D patients as high or low risk of DCVD. For the same set of SNP markers, the GLT and polygenic risk score (PRS) models showed similar predictive performance, and we observed consistency between the GLT and PRS models in that the model based on a larger number of SNP markers showed much-improved predictability. In silico gene expression analysis, additional information was provided on the functional role of the genes identified in this study. In particular, HDAC4, CDKN2B, CELSR2, and MRAS appear to be major hubs in the functional gene network for DCVD. The proposed risk prediction approach based on the liability threshold model may help identify T2D patients at high CVD risk in East Asian populations with further external validations.

How to Utilize Clinical and Genetic Information for Personalized Treatment of Major Depressive Disorder: Step by Step Strategic Approach

Depression is the single largest contributor to non-fatal health loss and affects 322 million people globally. The clinical heterogeneity of this disorder shows biological correlates and it makes the personalization of antidepressant prescription an important pillar of treatment. There is increasing evidence of genetic overlap between depression, other psychiatric and non-psychiatric disorders, which varies across depression subtypes. Therefore, the first step of clinical evaluation should include a careful assessment of psychopathology and physical health, not limited to previously diagnosed disorders. In part of the patients indeed the pathogenesis of depression may be strictly linked to inflammatory and metabolic abnormalities, and the treatment should target these as much as the depressive symptoms themselves. When the evaluation of the symptom and drug tolerability profile, the concomitant biochemical abnormalities and physical conditions is not enough and at least one pharmacotherapy failed, the genotyping of variants in CYP2D6/CYP2C19 (cytochromes responsible for antidepressant metabolism) should be considered. Individuals with altered metabolism through one of these enzymes may benefit from some antidepressants rather than others or need dose adjustments. Finally, if available, the polygenic predisposition towards cardio-metabolic disorders can be integrated with non-genetic risk factors to tune the identification of patients who should avoid medications associated with this type of side effects. A sufficient knowledge of the polygenic risk of complex medical and psychiatric conditions is becoming relevant as this information can be obtained through direct-to-consumer genetic tests and in the future it may provided by national health care systems.

Risk Stratification for Bipolar Disorder Using Polygenic Risk Scores Among Young High-Risk Adults

Objective: Identifying high-risk groups with an increased genetic liability for bipolar disorder (BD) will provide insights into the etiology of BD and contribute to early detection of BD. We used the BD polygenic risk score (PRS) derived from BD genome-wide association studies (GWAS) to explore how such genetic risk manifests in young, high-risk adults. We postulated that BD-PRS would be associated with risk factors for BD. Methods: A final sample of 185 young, high-risk German adults (aged 18-35 years) were grouped into three risk groups and compared to a healthy control group (n = 1,100). The risk groups comprised 117 cases with attention deficit hyperactivity disorder (ADHD), 45 with major depressive disorder (MDD), and 23 help-seeking adults with early recognition symptoms [ER: positive family history for BD, (sub)threshold affective symptomatology and/or mood swings, sleeping disorder]. BD-PRS was computed for each participant. Logistic regression models (controlling for sex, age, and the first five ancestry principal components) were used to assess associations of BD-PRS and the high-risk phenotypes. Results: We observed an association between BD-PRS and combined risk group status (OR = 1.48, p < 0.001), ADHD diagnosis (OR = 1.32, p = 0.009), MDD diagnosis (OR = 1.96, p < 0.001), and ER group status (OR = 1.7, p = 0.025; not significant after correction for multiple testing) compared to healthy controls. Conclusions: In the present study, increased genetic risk for BD was a significant predictor for MDD and ADHD status, but not for ER. These findings support an underlying shared risk for both MDD and BD as well as ADHD and BD. Improving our understanding of the underlying genetic architecture of these phenotypes may aid in early identification and risk stratification.

A cross-disorder PRS-pheWAS of 5 major psychiatric disorders in UK Biobank

Psychiatric disorders are highly heritable and associated with a wide variety of social adversity and physical health problems. Using genetic liability (rather than phenotypic measures of disease) as a proxy for psychiatric disease risk can be a useful alternative for research questions that would traditionally require large cohort studies with long-term follow up. Here we conducted a hypothesis-free phenome-wide association study in about 330,000 participants from the UK Biobank to examine associations of polygenic risk scores (PRS) for five psychiatric disorders (major depression (MDD), bipolar disorder (BP), schizophrenia (SCZ), attention-deficit/ hyperactivity disorder (ADHD) and autism spectrum disorder (ASD)) with 23,004 outcomes in UK Biobank, using the open-source PHESANT software package. There was evidence after multiple testing (p<2.55x10^-06) for associations of PRSs with 294 outcomes, most of them attributed to associations of PRS_MDD (n = 167) and PRS_SCZ (n = 157) with mental health factors. Among others, we found strong evidence of association of higher PRS_ADHD with 1.1 months younger age at first sexual intercourse [95% confidence interval [CI]: -1.25,-0.92] and a history of physical maltreatment; PRS_ASD with 0.01% lower erythrocyte distribution width [95%CI: -0.013,-0.007]; PRS_SCZ with 0.95 lower odds of playing computer games [95%CI:0.95,0.96]; PRS_MDD with a 0.12 points higher neuroticism score [95%CI:0.111,0.135] and PRS_BP with 1.03 higher odds of having a university degree [95%CI:1.02,1.03]. We were able to show that genetic liabilities for five major psychiatric disorders associate with long-term aspects of adult life, including socio-demographic factors, mental and physical health. This is evident even in individuals from the general population who do not necessarily present with a psychiatric disorder diagnosis.

Psychiatric disorders are associated with a wide range of adverse health, social and economic problems. Our study investigated the association of genetic risk for five common psychiatric disorders with socio- demographics, lifestyle and health of about 330,000 participants in the UK Biobank using a systematic, hypothesis-free approach. We found that genetic risk for attention deficit/hyperactivity disorder (ADHD) and bipolar disorder were most strongly associated with lifestyle factors, such as time of first sexual intercourse and educational attainment. Genetic risks for autism spectrum disorder and schizophrenia were associated with altered blood cell counts and decreased risk of playing computer games, respectively. Increased genetic risk for depression was associated with other mental health outcomes such as neuroticism and irritability. In general, our results suggest that genetic risk for psychiatric disorders associates with a range of health and lifestyle traits that were measured in adulthood, in individuals from the general population who do not necessarily present with a psychiatric disorder diagnosis. However, it is important to note that these associations are not necessary causal but can also represent genetic correlation or be influenced by other factors, such as socio-economic factors and selection into the cohort. The findings should inform future research using causally informative designs.

Polygenic Scores to Assess Atherosclerotic Cardiovascular Disease Risk: Clinical Perspectives and Basic Implications

An individual's susceptibility to atherosclerotic cardiovascular disease is influenced by numerous clinical and lifestyle factors, motivating the multifaceted approaches currently endorsed for primary and secondary cardiovascular disease prevention. With growing knowledge of the genetic basis of atherosclerotic cardiovascular disease-in particular, coronary artery disease-and its contribution to disease pathogenesis, there is increased interest in understanding the potential clinical utility of a genetic predictor that might further refine the assessment and management of atherosclerotic cardiovascular disease risk. Rapid scientific and technological advances have enabled widespread genotyping efforts and dynamic research in the field of coronary artery disease genetic risk prediction. In this review, we describe how genomic analyses of coronary artery disease have been leveraged to create polygenic risk scores. We then discuss evaluations of the clinical utility of these scores, pertinent mechanistic insights gleaned, and practical considerations relevant to the implementation of polygenic risk scores in the health care setting.

Impulsivity and Venturesomeness in an Adult ADHD Sample: Relation to Personality, Comorbidity, and Polygenic Risk

While impulsivity is a basic feature of attention-deficit/hyperactivity disorder (ADHD), no study explored the effect of different components of the Impulsiveness (Imp) and Venturesomeness (Vent) scale (IV7) on psychiatric comorbidities and an ADHD polygenic risk score (PRS). We used the IV7 self-report scale in an adult ADHD sample of 903 patients, 70% suffering from additional comorbid disorders, and in a subsample of 435 genotyped patients. Venturesomeness, unlike immediate Impulsivity, is not specific to ADHD. We consequently analyzed the influence of Imp and Vent also in the context of a PRS on psychiatric comorbidities of ADHD. Vent shows a distinctly different distribution of comorbidities, e.g., less anxiety and depression. PRS showed no effect on different ADHD comorbidities, but correlated with childhood hyperactivity. In a complementary analysis using principal component analysis with Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition ADHD criteria, revised NEO Personality Inventory, Imp, Vent, and PRS, we identified three ADHD subtypes. These are an impulsive-neurotic type, an adventurous-hyperactive type with a stronger genetic component, and an anxious-inattentive type. Our study thus suggests the importance of adventurousness and the differential consideration of impulsivity in ADHD. The genetic risk is distributed differently between these subtypes, which underlines the importance of clinically motivated subtyping. Impulsivity subtyping might give insights into the organization of comorbid disorders in ADHD and different genetic background.

The genetic etiology of eosinophilic esophagitis

Eosinophilic esophagitis (EoE) is a chronic allergic disease associated with marked mucosal eosinophil accumulation. Multiple studies have reported a strong familial component to EoE, with the presence of EoE increasing the risk for other family members with EoE. Epidemiologic studies support an important role for environmental risk factors as modulators of genetic risk. In a small percentage of cases, including patients who have Mendelian diseases with co-occurrent EoE, rare genetic variation with large effect sizes could mediate EoE and explain multigenerational incidence in families. Common genetic risk variants mediate genetic risk for the majority of patients with EoE. Across the 31 reported independent EoE risk loci (P < 10-5), most of the EoE risk variants are located in between genes (36.7%) or within the introns of genes (42.4%). Although some variants do change the amino acid sequence of genes (2.2%), only 3 of the 31 EoE risk loci harbor an amino acid-changing variant. Thus most EoE risk loci are outside of the coding regions of genes, suggesting a key role for gene regulation in patients with EoE, which is consistent with most other complex diseases.

Polygenic risk scores in psychiatry: Will they be useful for clinicians?

Major psychiatric disorders are heritable but they are genetically complex. This means that, with certain exceptions, single gene markers will not be helpful for diagnosis. However, we are learning more about the large number of gene variants that, in combination, are associated with risk for disorders such as schizophrenia, bipolar disorder, and other psychiatric conditions. The presence of those risk variants may now be combined into a polygenic risk score (PRS). Such a score provides a quantitative index of the genomic burden of risk variants in an individual, which relates to the likelihood that a person has a particular disorder. Currently, such scores are quite useful in research, and they are telling us much about the relationships between different disorders and other indices of brain function. In the future, as the datasets supporting the development of such scores become larger and more diverse and as methodological developments improve predictive capacity, we expect that PRS will have substantial clinical utility in the assessment of risk for disease, subtypes of disease, and even treatment response. Here, we provide an overview of PRS in general terms (including a glossary suitable for informed non-geneticists) and discuss the use of PRS in psychiatry, including their limitations and cautions for interpretation, as well as their applications now and in the future.

Progress in Polygenic Composite Scores in Alzheimer's and Other Complex Diseases

Advances in high-throughput genotyping and next-generation sequencing (NGS) coupled with larger sample sizes brings the realization of precision medicine closer than ever. Polygenic approaches incorporating the aggregate influence of multiple genetic variants can contribute to a better understanding of the genetic architecture of many complex diseases and facilitate patient stratification. This review addresses polygenic concepts, methodological developments, hypotheses, and key issues in study design. Polygenic risk scores (PRSs) have been applied to many complex diseases and here we focus on Alzheimer's disease (AD) as a primary exemplar. This review was designed to serve as a starting point for investigators wishing to use PRSs in their research and those interested in enhancing clinical study designs through enrichment strategies.

Machine Learning SNP Based Prediction for Precision Medicine

In the past decade, precision genomics based medicine has emerged to provide tailored and effective healthcare for patients depending upon their genetic features. Genome Wide Association Studies have also identified population based risk genetic variants for common and complex diseases. In order to meet the full promise of precision medicine, research is attempting to leverage our increasing genomic understanding and further develop personalized medical healthcare through ever more accurate disease risk prediction models. Polygenic risk scoring and machine learning are two primary approaches for disease risk prediction. Despite recent improvements, the results of polygenic risk scoring remain limited due to the approaches that are currently used. By contrast, machine learning algorithms have increased predictive abilities for complex disease risk. This increase in predictive abilities results from the ability of machine learning algorithms to handle multi-dimensional data. Here, we provide an overview of polygenic risk scoring and machine learning in complex disease risk prediction. We highlight recent machine learning application developments and describe how machine learning approaches can lead to improved complex disease prediction, which will help to incorporate genetic features into future personalized healthcare. Finally, we discuss how the future application of machine learning prediction models might help manage complex disease by providing tissue-specific targets for customized, preventive interventions.

Prediction of Schizophrenia Diagnosis by Integration of Genetically Correlated Conditions and Traits

Schizophrenia is genetically heterogeneous and comorbid with many conditions. In this study, we explored polygenic scores (PGSs) from genetically related conditions and traits to predict schizophrenia diagnosis using both logistic regression and deep neural network (DNN) models. We used the combined Molecular Genetics of Schizophrenia and Swedish Schizophrenia Case Control Study (MGS + SSCCS) data for training and testing the models, and used the Clinical Antipsychotic Trials for Intervention Effectiveness (CATIE) data as independent validation. We screened 28 conditions and traits comorbid with schizophrenia to identify traits as potential predictors and used LASSO regression to select predictors for model construction. We investigated how PGS calculation influenced model performance. We found that the inclusion of comorbid traits improved model performance and PGSs calculated from two traits were more generalizable in independent validation. With a DNN model using 19 PGS predictors, we accomplished a prediction accuracy of 0.813 and an AUC of 0.905 in the MGS + SSCCS data. When this model was validated with the CATIE data, it achieved an accuracy of 0.721 and AUC of 0.747. Our results indicate that PGSs alone may not be sufficient to predict schizophrenia accurately and the inclusion of behavioral and clinical data may be necessary for more accurate prediction model.

Replicability and Prediction: Lessons and Challenges from GWAS

Since the publication of the Wellcome Trust Case Control Consortium (WTCCC) landmark study a decade ago, genome-wide association studies (GWAS) have led to the discovery of thousands of risk variants involved in disease etiology. This success story has two angles that are often overlooked. First, GWAS findings are highly replicable. This is an unprecedented phenomenon in complex trait genetics, and indeed in many areas of science, which in past decades have been plagued by false positives. At a time of increasing concerns about the lack of reproducibility, we examine the biological and methodological reasons that account for the replicability of GWAS and identify the challenges ahead. In contrast to the exemplary success of disease gene discovery, at present GWAS findings are not useful for predicting phenotypes. We close with an overview of the prospects for individualized prediction of disease risk and its foreseeable impact in clinical practice.

Polygenic Risk Scores in Clinical Psychology: Bridging Genomic Risk to Individual Differences

Genomewide association studies (GWASs) across psychiatric phenotypes have shown that common genetic variants generally confer risk with small effect sizes (odds ratio < 1.1) that additively contribute to polygenic risk. Summary statistics derived from large discovery GWASs can be used to generate polygenic risk scores (PRS) in independent, target data sets to examine correlates of polygenic disorder liability (e.g., does genetic liability to schizophrenia predict cognition?). The intuitive appeal and generalizability of PRS have led to their widespread use and new insights into mechanisms of polygenic liability. However, when currently applied across traits they account for small amounts of variance (<3%), are relatively uninformative for clinical treatment, and, in isolation, provide no insight into molecular mechanisms. Larger GWASs are needed to increase the precision of PRS, and novel approaches integrating various data sources (e.g., multitrait analysis of GWASs) may improve the utility of current PRS.

Human Demographic History Impacts Genetic Risk Prediction across Diverse Populations

The vast majority of genome-wide association studies (GWASs) are performed in Europeans, and their transferability to other populations is dependent on many factors (e.g., linkage disequilibrium, allele frequencies, genetic architecture). As medical genomics studies become increasingly large and diverse, gaining insights into population history and consequently the transferability of disease risk measurement is critical. Here, we disentangle recent population history in the widely used 1000 Genomes Project reference panel, with an emphasis on populations underrepresented in medical studies. To examine the transferability of single-ancestry GWASs, we used published summary statistics to calculate polygenic risk scores for eight well-studied phenotypes. We identify directional inconsistencies in all scores; for example, height is predicted to decrease with genetic distance from Europeans, despite robust anthropological evidence that West Africans are as tall as Europeans on average. To gain deeper quantitative insights into GWAS transferability, we developed a complex trait coalescent-based simulation framework considering effects of polygenicity, causal allele frequency divergence, and heritability. As expected, correlations between true and inferred risk are typically highest in the population from which summary statistics were derived. We demonstrate that scores inferred from European GWASs are biased by genetic drift in other populations even when choosing the same causal variants and that biases in any direction are possible and unpredictable. This work cautions that summarizing findings from large-scale GWASs may have limited portability to other populations using standard approaches and highlights the need for generalized risk prediction methods and the inclusion of more diverse individuals in medical genomics.