Machine learning optimized polygenic scores for blood cell traits identify sex-specific trajectories and genetic correlations with disease

Genome-wide analyses of variance in blood cell phenotypes provide new insights into complex trait biology and prediction

Blood cell phenotypes are routinely tested in healthcare to inform clinical decisions. Genetic variants influencing mean blood cell phenotypes have been used to understand disease aetiology and improve prediction; however, additional information may be captured by genetic effects on observed variance. Here, we mapped variance quantitative trait loci (vQTL), i.e. genetic loci associated with trait variance, for 29 blood cell phenotypes from the UK Biobank (N~408,111). We discovered 176 independent blood cell vQTLs, of which 147 were not found by additive QTL mapping. vQTLs displayed on average 1.8-fold stronger negative selection than additive QTL, highlighting that selection acts to reduce extreme blood cell phenotypes. Variance polygenic scores (vPGSs) were constructed to stratify individuals in the INTERVAL cohort (N~40,466), where genetically less variable individuals (low vPGS) had increased conventional PGS accuracy (by ~19%) than genetically more variable individuals. Genetic prediction of blood cell traits improved by ~10% on average combining PGS with vPGS. Using Mendelian randomisation and vPGS association analyses, we found that alcohol consumption significantly increased blood cell trait variances highlighting the utility of blood cell vQTLs and vPGSs to provide novel insight into phenotype aetiology as well as improve prediction.

Immunogenetics of lithium response and psychiatric phenotypes in patients with bipolar disorder

The link between bipolar disorder (BP) and immune dysfunction remains controversial. While epidemiological studies have long suggested an association, recent research has found only limited evidence of such a relationship. To clarify this, we performed an exploratory study of the contributions of immune-relevant genetic factors to the response to lithium (Li) treatment and the clinical presentation of BP. First, we assessed the association of a large collection of immune-related genes (4925) with Li response, defined by the Retrospective Assessment of the Lithium Response Phenotype Scale (Alda scale), and clinical characteristics in patients with BP from the International Consortium on Lithium Genetics (ConLi+Gen, N = 2374). Second, we calculated here previously published polygenic scores (PGSs) for immune-related traits and evaluated their associations with Li response and clinical features. Overall, we observed relatively weak associations (p < 1 × 10-4) with BP phenotypes within immune-related genes. Network and functional enrichment analyses of the top findings from the association analyses of Li response variables showed an overrepresentation of pathways participating in cell adhesion and intercellular communication. These appeared to converge on the well-known Li-induced inhibition of GSK-3β. Association analyses of age-at-onset, number of mood episodes, and presence of psychosis, substance abuse and/or suicidal ideation suggested modest contributions of genes such as RTN4, XKR4, NRXN1, NRG1/3 and GRK5 to disease characteristics. PGS analyses returned weak associations (p < 0.05) between inflammation markers and the studied BP phenotypes. Our results suggest a modest relationship between immunity and clinical features in BP. More research is needed to assess the potential therapeutic relevance.

Highly parameterized polygenic scores tend to overfit to population stratification via random effects

Polygenic scores (PGSs), increasingly used in clinical settings, frequently include many genetic variants, with performance typically peaking at thousands of variants. Such highly parameterized PGSs often include variants that do not pass a genome-wide significance threshold. We propose a mathematical perspective that renders the effects of many of these non-significant variants random rather than causal, with the randomness capturing population structure. We devise methods to assess variant effect randomness and population stratification bias. Applying these methods to 141 traits from the UK Biobank, we find that, for many PGSs, the effects of non-significant variants are considerably random, with the extent of randomness associated with the degree of overfitting to population structure of the discovery cohort. Our findings explain why highly parameterized PGSs simultaneously have superior cohort-specific performance and limited generalizability, suggesting the critical need for variant randomness tests in PGS evaluation. Supporting code and a dashboard are available at https://github.com/songlab-cal/StratPGS.

Immunogenetics of lithium response and psychiatric phenotypes in patients with bipolar disorder

The link between bipolar disorder (BP) and immune dysfunction remains controversial. While epidemiological studies have long suggested an association, recent research has found only limited evidence of such a relationship. To clarify this, we investigated the contributions of immune-relevant genetic factors to the response to lithium (Li) treatment and the clinical presentation of BP. First, we assessed the association of a large collection of immune-related genes (4,925) with Li response, defined by the Retrospective Assessment of the Lithium Response Phenotype Scale (Alda scale), and clinical characteristics in patients with BP from the International Consortium on Lithium Genetics (ConLi+Gen, N = 2,374). Second, we calculated here previously published polygenic scores (PGSs) for immune-related traits and evaluated their associations with Li response and clinical features. We found several genes associated with Li response at p < 1×10- 4 values, including HAS3, CNTNAP5 and NFIB. Network and functional enrichment analyses uncovered an overrepresentation of pathways involved in cell adhesion and intercellular communication, which appear to converge on the well-known Li-induced inhibition of GSK-3β. We also found various genes associated with BP's age-at-onset, number of mood episodes, and presence of psychosis, substance abuse and/or suicidal ideation at the exploratory threshold. These included RTN4, XKR4, NRXN1, NRG1/3 and GRK5. Additionally, PGS analyses suggested serum FAS, ECP, TRANCE and cytokine ligands, amongst others, might represent potential circulating biomarkers of Li response and clinical presentation. Taken together, our results support the notion of a relatively weak association between immunity and clinically relevant features of BP at the genetic level.

An Optimal Stacked ResNet-BiLSTM-Based Accurate Detection and Classification of Genetic Disorders

Gene is located inside the nuclease and the genetic data is contained in deoxyribonucleic acid (DNA). A person's gene count ranges from 20,000 to 30,000. Even a minor alteration to the DNA sequence can be harmful if it affects the cell's fundamental functions. As a result, the gene begins to act abnormally. The sorts of genetic abnormalities brought on by mutation include chromosomal disorders, complex disorders, and single-gene disorders. Therefore, a detailed diagnosis method is required. Thus, we proposed an Elephant Herd Optimization-Whale Optimization Algorithm (EHO-WOA) optimized Stacked ResNet-Bidirectional Long Term Short Memory (ResNet-BiLSTM) model for detecting genetic disorders. Here, a hybrid EHO-WOA algorithm is presented to assess the Stacked ResNet-BiLSTM architecture's fitness. The ResNet-BiLSTM design uses the genotype and gene expression phenotype as input data. Furthermore, the proposed method identifies rare genetic disorders such as Angelman Syndrome, Rett Syndrome, and Prader-Willi Syndrome. It demonstrates the effectiveness of the developed model with greater accuracy, recall, specificity, precision, and f1-score. Thus, a wide range of DNA deficiencies including Prader-Willi syndrome, Marfan syndrome, Early Onset Morbid Obesity, Rett syndrome, and Angelman syndrome are predicted accurately.

An atlas of genetic scores to predict multi-omic traits

The use of omic modalities to dissect the molecular underpinnings of common diseases and traits is becoming increasingly common. But multi-omic traits can be genetically predicted, which enables highly cost-effective and powerful analyses for studies that do not have multi-omics1. Here we examine a large cohort (the INTERVAL study2; n = 50,000 participants) with extensive multi-omic data for plasma proteomics (SomaScan, n = 3,175; Olink, n = 4,822), plasma metabolomics (Metabolon HD4, n = 8,153), serum metabolomics (Nightingale, n = 37,359) and whole-blood Illumina RNA sequencing (n = 4,136), and use machine learning to train genetic scores for 17,227 molecular traits, including 10,521 that reach Bonferroni-adjusted significance. We evaluate the performance of genetic scores through external validation across cohorts of individuals of European, Asian and African American ancestries. In addition, we show the utility of these multi-omic genetic scores by quantifying the genetic control of biological pathways and by generating a synthetic multi-omic dataset of the UK Biobank3 to identify disease associations using a phenome-wide scan. We highlight a series of biological insights with regard to genetic mechanisms in metabolism and canonical pathway associations with disease; for example, JAK-STAT signalling and coronary atherosclerosis. Finally, we develop a portal ( https://www.omicspred.org/ ) to facilitate public access to all genetic scores and validation results, as well as to serve as a platform for future extensions and enhancements of multi-omic genetic scores.

Discovery and systematic characterization of risk variants and genes for coronary artery disease in over a million participants

The discovery of genetic loci associated with complex diseases has outpaced the elucidation of mechanisms of disease pathogenesis. Here we conducted a genome-wide association study (GWAS) for coronary artery disease (CAD) comprising 181,522 cases among 1,165,690 participants of predominantly European ancestry. We detected 241 associations, including 30 new loci. Cross-ancestry meta-analysis with a Japanese GWAS yielded 38 additional new loci. We prioritized likely causal variants using functionally informed fine-mapping, yielding 42 associations with less than five variants in the 95% credible set. Similarity-based clustering suggested roles for early developmental processes, cell cycle signaling and vascular cell migration and proliferation in the pathogenesis of CAD. We prioritized 220 candidate causal genes, combining eight complementary approaches, including 123 supported by three or more approaches. Using CRISPR-Cas9, we experimentally validated the effect of an enhancer in MYO9B, which appears to mediate CAD risk by regulating vascular cell motility. Our analysis identifies and systematically characterizes >250 risk loci for CAD to inform experimental interrogation of putative causal mechanisms for CAD.

Polygenic score development in the era of large-scale biobanks

In this issue of Cell Genomics, Xu et al. report a comprehensive analysis of the genetics of 26 blood cell traits, leveraging data from two large biobanks to construct and make available machine-learning optimized polygenic scores (PGSs). In addition to delivering insights into the biology and clinical associations of these traits, the authors evaluate and provide recommendations on methods for PGS construction.

Comparison of Methods Utilizing Sex-Specific PRSs Derived From GWAS Summary Statistics

The polygenic risk score (PRS) is calculated as the weighted sum of an individual's genotypes and their estimated effect sizes, which is often used to estimate an individual's genetic susceptibility to complex traits and disorders. It is well known that some complex human traits or disorders have sex differences in trait distributions, disease onset, progression, and treatment response, although the underlying mechanisms causing these sex differences remain largely unknown. PRSs for these traits are often based on Genome-Wide Association Studies (GWAS) data with both male and female samples included, ignoring sex differences. In this study, we present a benchmark study using both simulations with various combinations of genetic correlation and sample size ratios between sexes and real data to investigate whether combining sex-specific PRSs can outperform sex-agnostic PRSs on traits showing sex differences. We consider two types of PRS models in our study: single-population PRS models (PRScs, LDpred2) and multiple-population PRS models (PRScsx). For each trait or disorder, the candidate PRSs were calculated based on sex-specific GWAS data and sex-agnostic GWAS data. The simulation results show that applying LDpred2 or PRScsx to sex-specific GWAS data and then combining sex-specific PRSs leads to the highest prediction accuracy when the genetic correlation between sexes is low and the sample sizes for both sexes are balanced and large. Otherwise, the PRS generated by applying LDpred2 or PRScs to sex-agnostic GWAS data is more appropriate. If the sample sizes between sexes are not too small and very unbalanced, combining LDpred2-based sex-specific PRSs to predict on the sex with a larger sample size and combining PRScsx-based sex-specific PRSs to predict on the sex with a smaller size are the preferred strategies. For real data, we considered 19 traits from Genetic Investigation of ANthropometric Traits (GIANT) consortium studies and UK Biobank with both sex-specific GWAS data and sex-agnostic GWAS data. We found that for waist-to-hip ratio (WHR) related traits, accounting for sex differences and incorporating information from the opposite sex could help improve PRS prediction accuracy. Taken together, our findings in this study provide guidance on how to calculate the best PRS for sex-differentiated traits or disorders, especially as the sample size of GWASs grows in the future.