Biology subjects

Matthew Stephens

Publications and source records attributed to Matthew Stephens.

13 recordsLinked to original sources

bioRxiv2016

Estimating time to the common ancestor for a beneficial allele

The haplotypes of a beneficial allele carry information about its history that can shed light on its age and putative cause for its increase in frequency. Specifically, the signature of an alleles age is contained in the pattern of local ancestry that mutation and recombination impose on its haplotypic background. We provide a method to exploit this pattern and infer the time to the common ancestor of a positively selected allele following a rapid increase in frequency. We do so using a hidden Markov model which leverages the length distribution of the shared ancestral haplotype, the accumulation of derived mutations on the ancestral background, and the surrounding background haplotype diversity. Using simulations, we demonstrate how the inclusion of information from both mutation and recombination events increases accuracy relative to approaches that only consider a single type of event. We also show the behavior of the estimator in cases where data do not conform to model assumptions, and provide some diagnostics for assessing and improving inference. Using the method, we analyze population-specific patterns in the 1000 Genomes Project data to provide a global perspective on the timing of adaptation for several variants which show evidence of recent selection and functional relevance to diet, skin pigmentation, and morphology in humans.

Evolutionary Biology↗

bioRxiv2016

Clustering RNA-seq expression data using grade of membership models

Grade of membership models, also known as \"admixture models\", \"topic models\" or \"Latent Dirichlet Allocation\", are a generalization of cluster models that allow each sample to have membership in multiple clusters. These models are widely used in population genetics to model admixed individuals who have ancestry from multiple \"populations\", and in natural language processing to model documents having words from multiple \"topics\". Here we illustrate the potential for these models to cluster samples of RNA-seq gene expression data, measured on either bulk samples or single cells. We also provide methods to help interpret the clusters, by identifying genes that are distinctively expressed in each cluster. By applying these methods to several example RNA-seq applications we demonstrate their utility in identifying and summarizing structure and heterogeneity. Applied to data from the GTEx project on 53 human tissues, the approach highlights similarities among biologically-related tissues and identifies distinctively-expressed genes that recapitulate known biology. Applied to single-cell expression data from mouse preimplantation embryos, the approach highlights both discrete and continuous variation through early embryonic development stages, and highlights genes involved in a variety of relevant processes - from germ cell development, through compaction and morula formation, to the formation of inner cell mass and trophoblast at the blastocyst stage. The methods are implemented in the Bioconductor package CountClust.\n\nAuthor SummaryGene expression profile of a biological sample (either from single cells or pooled cells) results from a complex interplay of multiple related biological processes. Consequently, for example, distal tissue samples may share a similar gene expression profile through some common underlying biological processes. Our goal here is to illustrate that grade of membership (GoM) models - an approach widely used in population genetics to cluster admixed individuals who have ancestry from multiple populations - provide an attractive approach for clustering biological samples of RNA sequencing data. The GoM model allows each biological sample to have partial memberships in multiple biologically-distinct clusters, in contrast to traditional clustering methods that partition samples into distinct subgroups. We also provide methods for identifying genes that are distinctively expressed in each cluster to help biologically interpret the results. Applied to a dataset of 53 human tissues, the GoM approach highlights similarities among biologically-related tissues and identifies distinctively-expressed genes that recapitulate known biology. Applied to gene expression data of single cells from mouse preimplantation embryos, the approach highlights both discrete and continuous variation through early embryonic development stages, and genes involved in a variety of relevant processes. Our study highlights the potential of GoM models for elucidating biological structure in RNA-seq gene expression data.

Genetics↗

bioRxiv2016

Variance Adaptive Shrinkage (vash): Flexible Empirical Bayes estimation of variances

MotivationWe consider the problem of estimating variances on a large number of \"similar\" units, when there are relatively few observations on each unit. This problem is important in genomics, for example, where it is often desired to estimate variances for thousands of genes (or some other genomic unit) from just a few measurements on each. A common approach to this problem is to use an Empirical Bayes (EB) method that assumes the variances among genes follow an inverse-gamma distribution. Here we describe a more flexible EB method, whose main assumption is that the distribution of the variances (or, as an alternative, the precisions) is unimodal.\n\nResultsWe show that this more flexible assumption provides competitive performance with existing methods when the variances truly come from an inverse-gamma distribution, and can outperform them when the distribution of the variances is more complex. In analyses of several human gene expression datasets from the Genotype Tissues Expression (GTEx) consortium, we find that our more flexible model often fits the data appreciably better than the single inverse gamma distribution. At the same time we find that, for variance estimation, the differences between methods is often small, suggesting that the simpler methods will often suffice in practice.\n\nAvailabilityOur methods are implemented in an R package vashr available from http://github.com/mengyin/vashr.

Bioinformatics↗

bioRxiv2016

Bayesian large-scale multiple regression with summary statistics from genome-wide association studies

Bayesian methods for large-scale multiple regression provide attractive approaches to the analysis of genome-wide association studies (GWAS). For example, they can estimate heritability of complex traits, allowing for both polygenic and sparse models; and by incorporating external genomic data into the priors they can increase power and yield new biological insights. However, these methods require access to individual genotypes and phenotypes, which are often not easily available. Here we provide a framework for performing these analyses without individual-level data. Specifically, we introduce a \"Regression with Summary Statistics\" (RSS) likelihood, which relates the multiple regression coefficients to univariate regression results that are often easily available. The RSS likelihood requires estimates of correlations among covariates (SNPs), which also can be obtained from public databases. We perform Bayesian multiple regression analysis by combining the RSS likelihood with previously-proposed prior distributions, sampling posteriors by Markov chain Monte Carlo. In a wide range of simulations RSS performs similarly to analyses using the individual data, both for estimating heritability and detecting associations. We apply RSS to a GWAS of human height that contains 253,288 individuals typed at 1.06 million SNPs, for which analyses of individual-level data are practically impossible. Estimates of heritability (52%) are consistent with, but more precise, than previous results using subsets of these data. We also identify many previously-unreported loci that show evidence for association with height in our analyses. Software is available at https://github.com/stephenslab/rss.

Genetics↗

bioRxiv2016

False Discovery Rates: A New Deal.

We introduce a new Empirical Bayes approach for large-scale hypothesis testing, including estimating False Discovery Rates (FDRs), and effect sizes. This approach has two key differences from existing approaches to FDR analysis. First, it assumes that the distribution of the actual (unobserved) effects is unimodal, with a mode at 0. This \"unimodal assumption\" (UA), although natural in many contexts, is not usually incorporated into standard FDR analysis, and we demonstrate how incorporating it brings many benefits. Specifically, the UA facilitates efficient and robust computation - estimating the unimodal distribution involves solving a simple convex optimization problem - and enables more accurate inferences provided that it holds. Second, the method takes as its input two numbers for each test (an effect size estimate, and corresponding standard error), rather than the one number usually used (p value, or z score). When available, using two numbers instead of one helps account for variation in measurement precision across tests. It also facilitates estimation of effects, and unlike standard FDR methods our approach provides interval estimates (credible regions) for each effect in addition to measures of significance. To provide a bridge between interval estimates and significance measures we introduce the term \"local false sign rate\" to refer to the probability of getting the sign of an effect wrong, and argue that it is a superior measure of significance than the local FDR because it is both more generally applicable, and can be more robustly estimated. Our methods are implemented in an R package ashr available from http://github.com/stephens999/ashr.

Bioinformatics↗

bioRxiv2015

Thousands of novel translated open reading frames in humans inferred by ribosome footprint profiling

Accurate annotation of protein coding regions is essential for understanding how genetic information is translated into biological functions. Here we describe riboHMM, a new method that uses ribosome footprint data along with gene expression and sequence information to accurately infer translated sequences. We applied our method to human lymphoblastoid cell lines and identified 7,273 previously unannotated coding sequences, including 2,442 translated upstream open reading frames. We observed an enrichment of harringtonine-treated ribosome footprints at the inferred initiation sites, validating many of the novel coding sequences. The novel sequences exhibit significant signatures of selective constraint in the reading frames of the inferred proteins, suggesting that many of these are functional. Nearly 40% of bicistronic transcripts showed significant negative correlation in the levels of translation of their two coding sequences, suggesting a key regulatory role for these novel translated sequences. Our work significantly expands the set of known coding regions in humans.

Genomics↗

bioRxiv2015

A simple model-based approach to inferring and visualizing cancer mutation signatures

Recent advances in sequencing technologies have enabled the production of massive amounts of data on somatic mutations from cancer genomes. These data have led to the detection of characteristic patterns of somatic mutations or \"mutation signatures\" at an unprecedented resolution, with the potential for new insights into the causes and mechanisms of tumorigenesis.\n\nHere we present new methods for modelling, identifying and visualizing such mutation signatures. Our methods greatly simplify mutation signature models compared with existing approaches, reducing the number of parameters by orders of magnitude even while increasing the contextual factors (e.g. the number of flanking bases) that are accounted for. This improves both sensitivity and robustness of inferred signatures. We also provide a new intuitive way to visualize the signatures, analogous to the use of sequence logos to visualize transcription factor binding sites.\n\nWe illustrate our new method on somatic mutation data from urothelial carcinoma of the upper urinary tract, and a larger dataset from 30 diverse cancer types. The results illustrate several important features of our methods, including the ability of our new visualization tool to clearly highlight the key features of each signature, the improved robustness of signature inferences from small sample sizes, and more detailed inference of signature characteristics such as strand biases and sequence context effects at the base two positions 5 to the mutated site.\n\nThe overall framework of our work is based on probabilistic models that are closely connected with \"mixed-membership models\" which are widely used in population genetic admixture analysis, and in machine learning for document clustering. We argue that recognizing these relationships should help improve understanding of mutation signature extraction problems, and suggests ways to further improve the statistical methods.\n\nOur methods are implemented in an R package pmsignature (https://github.com/friend1ws/pmsignature) and a web application available at https://friend1ws.shinyapps.io/pmsignature_shiny/.\n\nAuthor SummarySomatic (non-inherited) mutations are acquired throughout our lives in cells throughout our body. These mutations can be caused, for example, by DNA replication errors or exposure to environmental mutagens such as tobacco smoke. Some of these mutations can lead to cancer.\n\nDifferent cancers, and even different instances of the same cancer, can show different distinctive patterns of somatic mutations. These distinctive patterns have become known as \"mutation signatures\". For example, C > A mutations are frequent in lung caners whereas C > T and CC > TT mutations are frequent in skin cancers. Each mutation signature may be associated with a specific kind of carcinogen, such as tobacco smoke or ultraviolet light. Identifying mutation signatures therefore has the potential to identify new carcinogens, and yield new insights into the mechanisms and causes of cancer,\n\nIn this paper, we introduce new statistical tools for tackling this important problem. These tools provide more robust and interpretable mutation signatures compared to previous approaches, as we demonstrate by applying them to large-scale cancer genomic data.

Bioinformatics↗

bioRxiv2014

msCentipede: Modeling heterogeneity across genomic sites improves accuracy in the inference of transcription factor binding

MotivationUnderstanding global gene regulation depends critically on accurate annotation of regulatory elements that are functional in a given cell type. CENTIPEDE, a powerful, probabilistic framework for identifying transcription factor binding sites from tissue-specific DNase I cleavage patterns and genomic sequence content, leverages the hypersensitivity of factor-bound chromatin and the information in the DNase I spatial cleavage profile characteristic of each DNA binding protein to accurately infer functional factor binding sites. However, the model for the spatial profile in this framework underestimates the substantial variation in the DNase I cleavage profiles across factor-bound genomic locations and across replicate measurements of chromatin accessibility.\n\nResultsIn this work, we adapt a multi-scale modeling framework for inhomogeneous Poisson processes to better model the underlying variation in DNase I cleavage patterns across genomic locations bound by a transcription factor. In addition to modeling variation, we also model spatial structure in the heterogeneity in DNase I cleavage patterns for each factor. Using DNase-seq measurements assayed in a lymphoblastoid cell line, we demonstrate the improved performance of this model for several transcription factors by comparing against the Chip-Seq peaks for those factors. Finally, we propose an extension to this framework that allows for a more flexible background model and evaluate the additional gain in accuracy achieved when the background model parameters are estimated using DNase-seq data from naked DNA. The proposed model can also be applied to paired-end ATAC-seq and DNase-seq data in a straightforward manner.\n\nAvailabilitymsCentipede, a Python implementation of an algorithm to infer transcription factor binding using this model, is made available at https://github.com/rajanil/msCentipede

Genomics↗

bioRxiv2014

Visualizing spatial population structure with estimated effective migration surfaces

Genetic data often exhibit patterns that are broadly consistent with \"isolation by distance\" - a phenomenon where genetic similarity tends to decay with geographic distance. In a heterogeneous habitat, decay may occur more quickly in some regions than others: for example, barriers to gene flow can accelerate the genetic differentiation between groups located close in space. We use the concept of \"effective migration\" to model the relationship between genetics and geography: in this paradigm, effective migration is low in regions where genetic similarity decays quickly. We present a method to quantify and visualize variation in effective migration across the habitat, which can be used to identify potential barriers to gene flow, from geographically indexed large-scale genetic data. Our approach uses a population genetic model to relate underlying migration rates to expected pairwise genetic dissimilarities, and estimates migration rates by matching these expectations to the observed dissimilarities. We illustrate the potential and limitations of our method using simulations and geo-referenced genetic data from elephant, human and Arabidopsis thaliana populations. The resulting visualizations highlight important features of the spatial population structure that are difficult to discern using existing methods for summarizing genetic variation such as principal components analysis.

Genetics↗

bioRxiv2014

A multivariate genome-wide association analysis of 10 LDL subfractions, and their response to statin treatment, in 1868 Caucasians

We conducted a genome-wide association analysis of 7 subfractions of low density lipoproteins (LDLs) and 3 subfractions of intermediate density lipoproteins (IDLs) measured by gradient gel electrophoresis, and their response to statin treatment, in 1868 individuals of European ancestry from the Pharmacogenomics and Risk of Cardiovascular Disease study. Our analyses identified four previously-implicated loci (SORT1, APOE, LPA, and CETP) as containing variants that are very strongly associated with lipoprotein subfractions (log10Bayes Factor > 15). Subsequent conditional analyses suggest that three of these (APOE, LPA and CETP) likely harbor multiple independently associated SNPs. Further, while different variants typically showed different characteristic patterns of association with combinations of subfractions, the two SNPs in CETP show strikingly similar patterns - both in our original data and in a replication cohort - consistent with a common underlying molecular mechanism. Notably, the CETP variants are very strongly associated with LDL subfractions, despite showing no association with total LDLs in our study, illustrating the potential value of the more detailed phenotypic measurements. In contrast with these strong subfraction associations, genetic association analysis of subfraction response to statins showed much weaker signals (none exceeding log10 Bayes Factor of 6). However, two SNPs (in APOE and LPA) previously-reported to be associated with LDL statin response do show some modest evidence for association in our data, and the subfraction response profiles at the LPA SNP are consistent with the LPA association, with response likely being due primarily to resistance of Lp(a) particles to statin therapy. An additional important feature of our analysis is that, unlike most previous analyses of multiple related phenotypes, we analyzed the subfractions jointly, rather than one at a time. Comparisons of our multivariate analyses with standard univariate analyses demonstrate that multivariate analyses can substantially increase power to detect associations. Software implementing our multivariate analysis methods is available at http://stephenslab.uchicago.edu/software.html\n\nAuthor SummaryLevels of plasma lipids and lipoproteins are related to risk of cardiovascular disease (CVD), and because of this, considerable attention has been devoted to genetic association analyses of lipid-related measures. In addition, motivated by the fact that statins are widely prescribed to lower plasma low density lipoprotein (LDL) cholesterol and CVD risk, and that response to statins has a genetic component, several studies have searched for genetic associations with response of lipid related phenotypes to statin treatment. Here, in 1868 individuals of European ancestry from the Pharmacogenomics and Risk of Cardiovascular Disease study, we have conducted genetic association analyses of 7 subfractions of LDLs and 3 subfractions of intermediate density lipoproteins (IDLs) measured by gradient gel electrophoresis, and their response to statin treatment. These phenotypic measurements offer higher resolution information on LDLs and IDLs than available previously. Therefore, our study provides a more detailed picture of association with the entire IDL/LDL subfraction profile than any prior genetic association studies of either lipid-related measures or their response to statin treatment. Moreover, unlike most previous analyses of multiple related measurements, we analyzed the subfractions jointly, rather than one at a time. Our results demonstrate that joint analyses of related measurements can considerably increase power to detect associations compared with conventional univariate analyses.

Genetics↗

bioRxiv2014

The Genetic Architecture of Gene Expression Levels in Wild Baboons

Gene expression variation is well documented in human populations and its genetic architecture has been extensively explored. However, we still know little about the genetic architecture of gene expression variation in other species, particularly our closest living relatives, the nonhuman primates. To address this gap, we performed an RNA sequencing (RNA-seq)-based study of 63 wild baboons, members of the intensively studied Amboseli baboon population in Kenya. Our study design allowed us to measure gene expression levels and identify genetic variants using the same data set, enabling us to perform complementary mapping of putative cis-acting expression quantitative trait loci (eQTL) and measurements of allele-specific expression (ASE) levels. We discovered substantial evidence for genetic effects on gene expression levels in this population. Surprisingly, we found more power to detect individual eQTL in the baboons relative to a HapMap human data set of comparable size, probably as a result of greater genetic variation, enrichment of SNPs with high minor allele frequencies, and longer-range linkage disequilibrium in the baboons. eQTL were most likely to be identified for lineage-specific, rapidly evolving genes. Interestingly, genes with eQTL significantly overlapped between the baboon and human data sets, suggesting that some genes may tolerate more genetic perturbation than others, and that this property may be conserved across species. Finally, we used a Bayesian sparse linear mixed model to partition genetic, demographic, and early environmental contributions to variation in gene expression levels. We found a strong genetic contribution to gene expression levels for almost all genes, while individual demographic and environmental effects tended to be more modest. Together, our results establish the feasibility of eQTL mapping using RNA-seq data alone, and act as an important first step towards understanding the genetic architecture of gene expression variation in nonhuman primates.

Genomics↗

bioRxiv2014

Epigenetic Modifications are Associated with Inter-species Gene Expression Variation in Primates

Changes in gene regulation level have long been thought to play an important role in evolution and speciation, especially in primates. Over the past decade, comparative genomic studies have revealed extensive inter-species differences in gene expression levels yet we know much less about the extent to which regulatory mechanisms differ between species. To begin addressing this gap, we performed a comparative epigenetic study in primate lymphoblastoid cell lines (LCLs), to query the contribution of RNA polymerase II (Pol II) and four histone modifications (H3K4me1, H3K4me3, H3K27ac, and H3K27me3) to inter-species variation in gene expression levels. We found that inter-species differences in mark enrichment near transcription start sites are significantly more often associated with inter-species differences in the corresponding gene expression level than expected by chance alone. Interestingly, we also found that first-order interactions among the histone marks and Pol II do not markedly contribute to the degree of association between the marks and inter-species variation in gene expression levels, suggesting that the marginal effects of the five marks dominate this contribution.

Genomics↗

bioRxiv2013

Variational Inference of Population Structure in Large SNP Datasets

Tools for estimating population structure from genetic data are now used in a wide variety of applications in population genetics. However, inferring population structure in large modern data sets imposes severe computational challenges. Here, we develop efficient algorithms for approximate inference of the model underlying the STRUCTURE program using a variational Bayesian framework. Variational methods pose the problem of computing relevant posterior distributions as an optimization problem, allowing us to build on recent advances in optimization theory to develop fast inference tools. In addition, we propose useful heuristic scores to identify the number of populations represented in a dataset and a new hierarchical prior to detect weak population structure in the data. We test the variational algorithms on simulated data, and illustrate using genotype data from the CEPH-Human Genome Diversity Panel. The variational algorithms are almost two orders of magnitude faster than STRUCTURE and achieve accuracies comparable to those of ADMIXTURE. Furthermore, our results show that the heuristic scores for choosing model complexity provide a reasonable range of values for the number of populations represented in the data, with minimal bias towards detecting structure when it is very weak. Our algorithm, fastSTRUCTURE, is freely available online at http://pritchardlab.stanford.edu/structure.html.

Genetics↗