Genetic association studies have yielded a wealth of biologic discoveries. However, these have mostly analyzed one trait and one SNP at a time, thus failing to capture the underlying complexity of these datasets. Joint genotype-phenotype analyses of complex, high-dimensional datasets represent an important way to move beyond simple GWAS with great potential. The move to high-dimensional phenotypes will raise many new statistical problems. In this paper we address the central issue of missing phenotypes in studies with any level of relatedness between samples. We propose a multiple phenotype mixed model and use a computationally efficient variational Bayesian algorithm to fit the model. On a variety of simulated and real datasets from a range of organisms and trait types, we show that our method outperforms existing state-of-the-art methods from the statistics and machine learning literature and can boost signals of association.

The vast majority of human mutations have minor allele frequencies (MAF) under 1%, with the plurality observed only once (i.e., “singletons”). While Mendelian diseases are predominantly caused by rare alleles, their cumulative contribution to complex phenotypes remains largely unknown. We develop and rigorously validate an approach to jointly estimate the contribution of all alleles, including singletons, to phenotypic variation. We apply our approach to transcriptional regulation, an intermediate between genetic variation and complex disease. Using whole genome DNA and lymphoblastoid cell line RNA sequencing data from 360 European individuals, we conservatively estimate that singletons contribute ~25% of cis -heritability across genes (dwarfing the contributions of other frequencies). The majority (~76%) of singleton heritability derives from ultra-rare variants absent from thousands of additional samples. We develop a novel inference procedure to demonstrate that our results are consistent with pervasive purifying selection shaping the regulatory architecture of most human genes.

Circulating cell-free DNA (cfDNA) in the bloodstream originates from dying cells and is a promising noninvasive biomarker for cell death. Here, we propose an algorithm, CelFiE, to accurately estimate the relative abundances of cell types and tissues contributing to cfDNA from epigenetic cfDNA sequencing. In contrast to previous work, CelFiE accommodates low coverage data, does not require CpG site curation, and estimates contributions from multiple unknown cell types that are not available in external reference data. In simulations, CelFiE accurately estimates known and unknown cell type proportions from low coverage and noisy cfDNA mixtures, including from cell types composing less than 1% of the total mixture. When used in two clinically-relevant situations, CelFiE correctly estimates a large placenta component in pregnant women, and an elevated skeletal muscle component in amyotrophic lateral sclerosis (ALS) patients, consistent with the occurrence of muscle wasting typical in these patients. Together, these results show how CelFiE could be a useful tool for biomarker discovery and monitoring the progression of degenerative disease.

The genetic control of gene expression is a core component of human physiology. For the past several years, transcriptome-wide association studies have leveraged large datasets of linked genotype and RNA sequencing information to create a powerful gene-based test of association that has been used in dozens of studies. While numerous discoveries have been made, the populations in the training data are overwhelmingly of European descent, and little is known about the generalizability of these models to other populations. Here, we test for cross-population generalizability of gene expression prediction models using a dataset of African American individuals with RNA-Seq data in whole blood. We find that the default models trained in large datasets such as GTEx and DGN fare poorly in African Americans, with a notable reduction in prediction accuracy when compared to European Americans. We replicate these limitations in cross-population generalizability using the five populations in the GEUVADIS dataset. Via realistic simulations of both populations and gene expression, we show that accurate cross-population generalizability of transcriptome prediction only arises when eQTL architecture is substantially shared across populations. In contrast, models with non-identical eQTLs showed patterns similar to real-world data. Therefore, generating RNA-Seq data in diverse populations is a critical step towards multiethnic utility of gene expression prediction.