Schedule for: 24w5292 - Statistical, Computational, Translational, and Ethical Challenges in Biobank Data Analysis

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

A brief introduction to BIRS with important logistical information, technology instruction, and opportunity for participants to ask questions.

Biobank-linked electronic health record (EHR) is increasingly becoming available for genetic epidemiology and genomics research. This talk will discuss method development efforts to utilize multiple EHR datasets to identify cross-phenotype genetic associations and develop genetic risk prediction models. Results from the UK Biobank and the Electronic Medical Records and Genomics (eMERGE) Network will be presented to highlight the opportunities and challenges in using EHR data for genomics research.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

Meet in foyer of TCPL to participate in the BIRS group photo. The photograph will be taken outdoors, so dress appropriately for the weather. Please don't be late, or you might not be in the official group photo!

Online Mendelian Inheritance in Man (OMIM) lists 884 disease genes on the X chromosome, slightly more than chr 7 (860), the autosome with most similar physical length. In contrast, in 2013 Wise et al., showed that the X chr had only 12% of the number of GWAS signals compared to chr 7. Further, they showed that only 33% of GWAS reported including the X chromosome in 2010-2011. They made suggestions to improve this. Disappointingly, 10 years later we showed that the situation has become worse! In 2021 only 25% of GWAS summary statistics in the NHGRI-EBI GWAS catalog included results for the X (Sun et al., AJHG 2023). X chr variants are also typically absent from scores in the PGS Catalog. I will discuss 3 data quality issues that may be contributing to the reluctance to analyse X chr variants: sex differences in genotype missingness; sex differences in allele frequency; deviation from Hardy-Weinberg equilibrium. I provide examples of these phenomenon in large datasets. This is joint work with Lei Sun.

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

GWAS have identified numerous associations for complex traits. Most associated variants are non-coding which may function by regulating gene expression levels. Understanding the function of non-coding GWAS hits requires integrating eQTL datasets using methods such as transcriptome-wide association studies (TWAS). Most existing TWAS use bulk tissues and may not have the resolution to reveal cell-type specific target genes. Single-cell expression quantitative trait loci (sc-eQTL) datasets are emerging. The largest bulk- and sc-eQTL datasets are most conveniently available as summary statistics, but have not been broadly utilized in TWAS. In this talk, we will discuss methods for analyzing sc-eQTL summary statistics and integrating them with GWAS data to identify risk genes. Applying these methods to analyze multi-ancestry GWAS datasets for 14 autoimmune diseases. We also develop a cell type aware drug repurposing pipeline, which leverages EXPRESSO results to identify drug compounds that can reverse disease gene expressions in relevant cell types. Our results point to multiple drugs with therapeutic potentials, including metformin for type 1 diabetes, and vitamin K for ulcerative colitis. In summary, methods that can integrate sc-eQTL datasets with GWAS can inform cell type specific risk genes and improve the discovery of potential therapeutics.

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

Extensive whole-genome/exome sequencing (WGS/WES) data, combined with electronic health records (EHRs) from large-scale national and institutional biobanks, have provided unique opportunities to advance our understanding of the contributions of rare variants in both coding and noncoding regions of the genome to complex human traits. In the meantime, challenges remain in analyzing biobank WGS/WES studies to maximize the utility of these rich and extensive data. In this talk, I will discuss several statistical and computational approaches for the integrative analysis of biobank WGS/WES studies from recent works, including STAAR for functionally-informed association analysis by leveraging multi-faceted variant annotation data; MetaSTAAR for resource-efficient meta-analysis of sequencing data from multiple studies using sparse LD matrices; and MultiSTAAR for joint modeling of multiple traits to detect pleiotropic genes and regions. These methods account for population structure and sample relatedness and are scalable for analyzing biobank-scale cohorts, and their applications will be illustrated using ongoing population-based WGS/WES studies, including the Trans-Omics Precision Medicine Program (TOPMed) from the National Heart, Lung and Blood Institute, the UK Biobank and the All of Us Program from the National Institutes of Health, which have been collectively sequencing for about 1 million ancestrally-diverse genomes.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

I will discuss two recent methodological works from our group on Mendelian randomization (MR). The first is an “almost exact” approach to within-family MR, which the various potential sources of biases from a graphical viewpoint of the MR design. The second is an approach that leverages biobank/multi-omic data to learn hidden mediators of the genetic effects. Link to first paper: https://arxiv.org/abs/2208.14035 (joint work with Matt Tudball and George Davey Smith). Link to second paper: https://www.statslab.cam.ac.uk/~qz280/publication/pathgps/paper.pdf (joint work with Zijun Gao and Trevor Hastie, accepted by Biometrics).

The exploration of gene-environment (GxE) interactions as contributors to the phenotypic variations of complex traits has long been a subject of debate. The limited reporting of such interactions to date may be attributed to the inherent challenge of low statistical power in detecting GxE using existing methodologies. In this study, I will present a novel approach for detecting GxE interactions that draws parallels to the Mendelian randomization framework—a widely utilized method for inferring causal relationships between exposures and disease outcomes. Our novel approach offers enhanced statistical power compared to direct GxE testing methods in certain scenarios and exhibits robustness against issues related to population structure and phenotype measurements. I will illustrate the efficacy of this method by investigating interactions associated with smoking or alcohol consumption on serum lipids and blood pressure.

Mendelian randomization (MR) is one of the most popular methods used for causal inference in recent biomedical literature. MR has essential instrumental variable assumptions, which, when violated, can lead to biased results. Although various methods have been developed to obtain robust causal effect estimates in the presence of invalid genetic instruments, they require additional conditions to be fulfilled. In this talk, I will discuss how large biobank data may result in more frequent violations of MR assumptions from two perspectives. First, MR aiming to evaluate potential causal effects of polygenic exposures can be biased due to the majority of the genetic instruments being subject to horizontal pleiotropy. I will introduce our approach for performing sensitivity analysis of MR based upon the core gene hypothesis. Second, both linear and non-linear MR analyses can be subject to confounding due to selection bias. I will discuss potential alternatives to mitigate such confounding effects as well as cautions that should be taken when making causal interpretations of MR results.

Brain imaging and genomics are critical tools enabling characterization of the genetic basis of brain disorders. However, imaging large cohorts is expensive and may be unavailable for legacy datasets used for genome-wide association studies (GWASs). Using an integrated feature selection/aggregation model, we developed an image-mediated association study (IMAS), which utilizes borrowed imaging/genomics data to conduct association mapping in legacy GWAS cohorts. By leveraging the UK Biobank image-derived phenotypes (IDPs), the IMAS discovered genetic bases underlying four neuropsychiatric disorders and verified them by analyzing annotations, pathways, and expression quantitative trait loci (eQTLs). A cerebellar-mediated mechanism was identified to be common to the four disorders. Simulations show that, if the goal is identifying genetic risk, our IMAS is more powerful than a hypothetical protocol in which the imaging results were available in the GWAS dataset. This implies the feasibility of reanalyzing legacy GWAS datasets without conducting additional imaging, yielding cost savings for integrated analysis of genetics and imaging.

Instrumental variable methods provide useful tools for inferring causal effects in the presence of unmeasured confounding. To apply these methods with large-scale data sets, a major challenge is to find valid instruments from a possibly large candidate set. In practice, most of the candidate instruments are often not relevant for studying a particular exposure of interest. Moreover, not all relevant candidate instruments are valid as they may directly influence the outcome of interest. In this article, we propose a data-driven method for causal inference with many candidate instruments that addresses these two challenges simultaneously. A key component of our proposal involves using pseudo variables, known to be irrelevant, to remove variables from the original set that exhibit spurious correlations with the exposure. Synthetic data analyses show that the proposed method performs favourably compared to existing methods. We apply our method to a Mendelian randomization study estimating the effect of obesity on health-related quality of life.

Lunch is served daily between 11:30am and 1:30pm in the Vistas Dining Room, the top floor of the Sally Borden Building.

The increased use of spatially resolved transcriptomics provides new biological insights into disease mechanisms. However, the high cost and complexity of these methods are not only barriers to broad clinical adoption, they also limit the number of matched samples for multi-omics analysis. Consequently, there is an emerging need to develop methods to predict spatial gene expression from routinely collected histology images. In this talk, I will discuss a new approach GHIST, a deep learning-based framework that predicts spatial gene expression at single-cell resolution by leveraging subcellular spatial transcriptomics such as Xenium technology and synergistic relationships between multiple layers of biological information. We then applied GHIST to selected samples in The Cancer Genome Atlas Breast Cancer H&E histology images. This in silico single-cell gene expression dataset allows us to investigate the associations between genomic profiles and tumour ecosystem via various spatial metrics. We found that cell type colocalisation has potential association with the identified breast cancer GWAS SNPs. Such analysis demonstrates GHIST capacity to enrich existing datasets with a spatially resolved omics modality, paving the way for scalable multi-omics analysis and new biomarker discoveries.

Genome-wide association studies (GWAS) have discovered many associations between genetic variants and complex traits. However, interpreting GWAS results, such as pinpointing causal variants, understanding the interplay between traits, and characterizing gene-environment interactions, is complicated by linkage disequilibrium, because univariate regression models cannot account for correlation between variants. Additionally, the large number of genetic variants poses computational challenges and incurs a high burden of multiple testing. Probabilistic graphical models can effectively represent dependencies between variables and provide flexible approaches for data integration. Therefore, they can be very useful in tackling these challenges in statistical genetics. In this talk, I will present two probabilistic graphical models, along with efficient variational inference algorithms to facilitate the interpretation of GWAS results. The first method, SparsePro, integrates GWAS associations and functional annotations for prioritizing causal variants. SparsePro demonstrated improved performance in simulations and identified biologically relevant causal variants. The second method, SharePro, assesses whether two or more traits share the same genetic signals identified in GWAS. SharePro achieved improved power with well-controlled false positive rate and identified biologically plausible colocalizations missed by existing methods. SharePro could be further adapted for gene-environment interaction analysis by accounting for genetic effect heterogeneity and could effectively reduce multiple testing burden. These new methods serve as valuable tools for improving our understanding of complex trait genetics.

Polygenic risk scores (PRS) hold prognostic value for identifying individuals at higher risk of chronic disease. However, further characterization is needed to understand the generalizability of PRS in diverse populations across various contexts. Here, we showed that lipids and T2D PRS performance is context-dependent by characterizing multi-ancestry genome-wide PRS among 600,000-800,000 participants across eight populations from the Population Architecture Genomics and Epidemiology (PAGE) Study and 13 additional biobanks and cohorts. We find that PRS performs better in younger individuals and those with family history of the trait. Performance per sex differs by trait. Additionally, the T2D PRS was associated with various conditions, particularly cardiometabolic-related traits and complications, suggesting its utility beyond T2D risk prediction and shared genetic architecture between T2D and other diseases. These findings highlight the need to account for contexts in PRS, with notable implications for the prognostic value and clinical utility of PRS.

Biobank data encompass a wealth of information on a large number of individuals, including extensive clinical phenotypes, genotyping, biomarkers, and various omics data. Typically, these individuals are not selected based on phenotypic status. While such data may not be ideal for studying rare diseases like cancer due to limited power, they provide valuable insights into the interrelationship between genotypes, biomarkers, and various molecular phenotypes. In this talk, I will describe a set-based association analysis that complements individual-variant analysis in genome-wide association studies (GWAS). This method leverages relationships derived from biobanks or other reference databases to study genetic associations with disease traits by aggregating variants whose signals are weak individually but significant when considered jointly with molecular phenotypes and biomarkers. We employ a unified mixed-effects model to formulate the overall association through the fixed effect of genetically predicted intermediate phenotypes and the random effects of individual variants. Our set-based score testing framework, MiST (Mixed-effects Score Test), includes data-driven combination approaches to jointly test for both fixed and random effects. We extend MiST to rely solely on GWAS summary statistics, which are widely available. Extensive simulations and real data analyses demonstrate that MiST is powerful, and the summary statistics-based MiST (sMiST) produces results that are consistent with those obtained from individual-level data, with significantly improved computational speed.

A buffet dinner is served daily between 5:30pm and 7:30pm in Vistas Dining Room, top floor of the Sally Borden Building.

Breakfast is served daily between 7 and 9am in the Vistas Dining Room, the top floor of the Sally Borden Building.

Approaches systematically characterizing interactions via transcriptomic data usually follow two systems: (i) coexpression network analyses focusing on correlations between genes and (ii) linear regressions (usually regularized) to select multiple genes jointly. Both suffer from the problem of stability: A slight change of parameterization or dataset could lead to marked alterations of outcomes. Here, we propose Stabilized COre gene and Pathway Election (SCOPE), a tool integrating bootstrapped least absolute shrinkage and selection operator and coexpression analysis, leading to robust outcomes insensitive to variations in data. By applying SCOPE to six cancer expression datasets (BRCA, COAD, KIRC, LUAD, PRAD, and THCA) in The Cancer Genome Atlas, we identified core genes capturing interaction effects in crucial pan-cancer pathways related to genome instability and DNA damage response. Moreover, we highlighted the pivotal role of CD63 as an oncogenic driver and a potential therapeutic target in kidney cancer. SCOPE enables stabilized investigations toward complex interactions using transcriptome data.

Background: Observational studies have found that vitamin D supplementation is associated with improved cognition. Further, recent Mendelian randomization (MR) studies have shown that higher vitamin D levels, 25(OH)D, may protect against Alzheimer's disease. Thus, it is possible that 25(OH)D may protect against Alzheimer's disease by improving cognition. Objective: We assessed this hypothesis, by examining the relationship between 25(OH)D levels and seven cognitive measurements. Methods: To mitigate bias from confounding, we performed two-sample MR analyses. We used instruments from three publications: Manousaki et al. (2020), Sutherland et al. (2022), and the Emerging Risk Factors Collaboration/EPIC-CVD/Vitamin D Studies Collaboration (2021). Results: Our observational studies suggested a protective association between 25(OH)D levels and cognitive measures. An increase in the natural log of 25(OH)D by 1 SD was associated with a higher PACC score (BetaPACC score = 0.06, 95% CI = (0.04-0.08); p = 1.8×10-10). However, in the MR analyses, the estimated effect of 25(OH)D on cognitive measures was null. Specifically, per 1 SD increase in genetically estimated natural log of 25(OH)D, the PACC scores remained unchanged in the overall population, (BetaPACC score = -0.01, 95% CI (-0.06 to 0.03); p = 0.53), and amongst individuals aged over 60 (BetaPACC score = 0.03, 95% CI (-0.028 to 0.08); p = 0.35). Conclusions: In conclusion, our MR study found no clear evidence to support a protective role of increased 25(OH)D concentrations on cognitive performance in European ancestry individuals. However, our study cannot entirely dismiss the potential beneficial effect on PACC for individuals over the age of 60.

5-day workshop participants are welcome to use BIRS facilities (TCPL ) until 3 pm on Friday, although participants are still required to checkout of the guest rooms by 11AM.