A Fast and Flexible Algorithm for Solving the Lasso in Large-scale and Ultrahigh-dimensional Problems, bioRxiv, 2019-05-07

AbstractSince its first proposal in statistics (Tibshirani, 1996), the lasso has been an effective method for simultaneous variable selection and estimation. A number of packages have been developed to solve the lasso efficiently. However as large datasets become more prevalent, many algorithms are constrained by efficiency or memory bounds. In this paper, we propose a meta algorithm batch screening iterative lasso (BASIL) that can take advantage of any existing lasso solver and build a scalable lasso solution for large datasets. We also introduce snpnet, an R package that implements the proposed algorithm on top of glmnet (Friedman et al., 2010a) for large-scale single nucleotide polymorphism (SNP) datasets that are widely studied in genetics. We demonstrate results on a large genotype-phenotype dataset from the UK Biobank, where we achieve state-of-the-art heritability estimation on quantitative and qualitative traits including height, body mass index, asthma and high cholesterol.

biorxiv bioinformatics 0-100-users 2019

Variable prediction accuracy of polygenic scores within an ancestry group, bioRxiv, 2019-05-07

AbstractFields as diverse as human genetics and sociology are increasingly using polygenic scores based on genome-wide association studies (GWAS) for phenotypic prediction. However, recent work has shown that polygenic scores have limited portability across groups of different genetic ancestries, restricting the contexts in which they can be used reliably and potentially creating serious inequities in future clinical applications. Using the UK Biobank data, we demonstrate that even within a single ancestry group, the prediction accuracy of polygenic scores depends on characteristics such as the age or sex composition of the individuals in which the GWAS and the prediction were conducted, and on the GWAS study design. Our findings highlight both the complexities of interpreting polygenic scores and underappreciated obstacles to their broad use.

biorxiv genetics 200-500-users 2019

Why are education, socioeconomic position and intelligence genetically correlated?, bioRxiv, 2019-05-07

AbstractGenetic associations and correlations are perceived as confirmation that genotype influences one or more phenotypes respectively. However, genetic correlations can arise from non-biological or indirect mechanisms including population stratification, dynastic effects, and assortative mating. In this paper, we outline these mechanisms and demonstrate available tools and analytic methods that can be used to assess their presence in estimates of genetic correlations and genetic associations. Using educational attainment and parental socioeconomic position data as an exemplar, we demonstrate that both heritability and genetic correlation estimates may be inflated by these indirect mechanisms. The results highlight the limitations of between-individual estimates obtained from samples of unrelated individuals and the potential value of family-based studies. Use of the highlighted tools in combination with within-sibling or mother-father-offspring trio data may offer researchers greater opportunity to explore the underlying mechanisms behind genetic associations and correlations and identify the underlying causes of estimate inflation.

biorxiv genetics 100-200-users 2019

Levels of Representation in a Deep Learning Model of Categorization, bioRxiv, 2019-05-06

AbstractDeep convolutional neural networks (DCNNs) rival humans in object recognition. The layers (or levels of representation) in DCNNs have been successfully aligned with processing stages along the ventral stream for visual processing. Here, we propose a model of concept learning that uses visual representations from these networks to build memory representations of novel categories, which may rely on the medial temporal lobe (MTL) and medial prefrontal cortex (mPFC). Our approach opens up two possibilities a) formal investigations can involve photographic stimuli as opposed to stimuli handcrafted and coded by the experimenter; b) model comparison can determine which level of representation within a DCNN a learner is using during categorization decisions. Pursuing the latter point, DCNNs suggest that the shape bias in children relies on representations at more advanced network layers whereas a learner that relied on lower network layers would display a color bias. These results confirm the role of natural statistics in the shape bias (i.e., shape is predictive of category membership) while highlighting that the type of statistics matter, i.e., those from lower or higher levels of representation. We use the same approach to provide evidence that pigeons performing seemingly sophisticated categorization of complex imagery may in fact be relying on representations that are very low-level (i.e., retinotopic). Although complex features, such as shape, relatively predominate at more advanced network layers, even simple features, such as spatial frequency and orientation, are better represented at the more advanced layers, contrary to a standard hierarchical view.

biorxiv neuroscience 100-200-users 2019

Long metabarcoding of the eukaryotic rDNA operon to phylogenetically and taxonomically resolve environmental diversity, bioRxiv, 2019-05-06

AbstractHigh-throughput environmental DNA metabarcoding has revolutionized the analysis of microbial diversity, but this approach is generally restricted to amplicon sizes below 500 base pairs. These short regions contain limited phylogenetic signal, which makes it impractical to use environmental DNA in full phylogenetic inferences. However, new long-read sequencing technologies such as the Pacific Biosciences platform may provide sufficiently large sequence lengths to overcome the poor phylogenetic resolution of short amplicons. To test this idea, we amplified soil DNA and used PacBio Circular Consensus Sequencing (CCS) to obtain a ~4500 bp region of the eukaryotic rDNA operon spanning most of the small (18S) and large subunit (28S) ribosomal RNA genes. The CCS reads were first treated with a novel curation workflow that generated 650 high-quality OTUs containing the physically linked 18S and 28S regions of the long amplicons. In order to assign taxonomy to these OTUs, we developed a phylogeny-aware approach based on the 18S region that showed greater accuracy and sensitivity than similarity-based and phylogenetic placement-based methods using shorter reads. The taxonomically-annotated OTUs were then combined with available 18S and 28S reference sequences to infer a well-resolved phylogeny spanning all major groups of eukaryotes, allowing to accurately derive the evolutionary origin of environmental diversity. A total of 1019 sequences were included, of which a majority (58%) corresponded to the new long environmental CCS reads. Comparisons to the 18S-only region of our amplicons revealed that the combined 18S-28S genes globally increased the phylogenetic resolution, recovering specific groupings otherwise missing. The long-reads also allowed to directly investigate the relationships among environmental sequences themselves, which represents a key advantage over the placement of short reads on a reference phylogeny. Altogether, our results show that long amplicons can be treated in a full phylogenetic framework to provide greater taxonomic resolution and a robust evolutionary perspective to environmental DNA.

biorxiv microbiology 100-200-users 2019

Screening human embryos for polygenic traits has limited utility, bioRxiv, 2019-05-06

AbstractGenome-wide association studies have led to the development of polygenic score (PS) predictors that explain increasing proportions of the variance in human complex traits. In parallel, progress in preimplantation genetic testing now allows genome-wide genotyping of embryos generated via in vitro fertilization (IVF). Jointly, these developments suggest the possibility of screening embryos for polygenic traits such as height or cognitive function. There are clear ethical, legal, and societal concerns regarding such a procedure, but these cannot be properly discussed in the absence of data on the expected outcomes of screening. Here, we use theory, simulations, and real data to evaluate the potential gain of PS-based embryo selection, defined as the expected difference in trait value between the top-scoring embryo and an average, unselected embryo. We observe that the gain increases very slowly with the number of embryos, but more rapidly with increased variance explained by the PS. Given currently available polygenic predictors and typical IVF yields, the average gain due to selection would be ≈2.5cm if selecting for height, and ≈2.5 IQ (intelligence quotient) points if selecting for cognitive function. These mean values are accompanied by wide confidence intervals; in real data drawn from nuclear families with up to 20 offspring each, we observe that the offspring with the highest PS for height was the tallest only in 25% of the families. We discuss prospects and limitations of PS-based embryo selection for the foreseeable future.

biorxiv genetics 200-500-users 2019