Search

Inverse reinforcement learning~(IRL) is a powerful framework to infer an agent's reward function by observing its behavior, but IRL algorithms that learn point estimates of the reward function can be misleading because there may be several functions that describe an agent's behavior equally well. A Bayesian approach to IRL models a distribution over candidate reward functions, alleviating the shortcomings of learning a point estimate. However, several Bayesian IRL algorithms use a $Q$-value function in place of the likelihood function. The resulting posterior is computationally intensive to calculate, has few theoretical guarantees, and the $Q$-value function is often a poor approximation for the likelihood. We introduce kernel density Bayesian IRL (KD-BIRL), which uses conditional kernel density estimation to directly approximate the likelihood, providing an efficient framework that, with a modified reward function parameterization, is applicable to environments with complex and infinite state spaces. We demonstrate KD-BIRL's benefits through a series of experiments in Gridworld environments and a simulated sepsis treatment task.

Gaussian processes are widely used for the analysis of spatial data due to their nonparametric flexibility and ability to quantify uncertainty, and recently developed scalable approximations have facilitated application to massive datasets. For multivariate outcomes, linear models of coregionalization combine dimension reduction with spatial correlation. However, their real-valued latent factors and loadings are difficult to interpret because, unlike nonnegative models, they do not recover a parts-based representation. We present nonnegative spatial factorization (NSF), a spatially-aware probabilistic dimension reduction model that naturally encourages sparsity. We compare NSF to real-valued spatial factorizations such as MEFISTO and nonspatial dimension reduction methods using simulations and high-dimensional spatial transcriptomics data. NSF identifies generalizable spatial patterns of gene expression. Since not all patterns of gene expression are spatial, we also propose a hybrid extension of NSF that combines spatial and nonspatial components, enabling quantification of spatial importance for both observations and features. A TensorFlow implementation of NSF is available from https://github.com/willtownes/nsf-paper .

Additional file 2 Supplemental figure 1–sensitivity of clustering results to initialization. The normalized mutual information and adjusted rand score when comparing clusters learned from K-Means applied to tGPLVM fit to true cluster labels when tGPLVM is initialized with PCA or random Gaussian noise.

Additional file 1 Appendix.

Additional file 1 A PDF file with two appendices. Appendix A describes the variational distributions for inference. Appendix B lists the sources of the various data sets and implementations of comparison methods.

The Genotype-Tissue Expression (GTEx) project was established to characterize genetic effects on the transcriptome across human tissues, and to link these regulatory mechanisms to trait and disease associations. Here, we present analyses of the v8 data, based on 17,382 RNA-sequencing samples from 54 tissues of 948 post-mortem donors. We comprehensively characterize genetic associations for gene expression and splicing incisandtrans, showing that regulatory associations are found for almost all genes, and describe the underlying molecular mechanisms and their contribution to allelic heterogeneity and pleiotropy of complex traits. Leveraging the large diversity of tissues, we provide insights into the tissue-specificity of genetic effects, and show that cell type composition is a key factor in understanding gene regulatory mechanisms in human tissues.

The problem of maximizing cell type discovery under budget constraints is a fundamental challenge for the collection and analysis of single-cell RNA-sequencing (scRNA-seq) data. In this paper, we introduce a simple, computationally efficient, and scalable Bayesian nonparametric sequential approach to optimize the budget allocation when designing a large scale experiment for the collection of scRNA-seq data for the purpose of, but not limited to, creating cell atlases. Our approach relies on the following tools: i) a hierarchical Pitman-Yor prior that recapitulates biological assumptions regarding cellular differentiation, and ii) a Thompson sampling multi-armed bandit strategy that balances exploitation and exploration to prioritize experiments across a sequence of trials. Posterior inference is performed by using a sequential Monte Carlo approach, which allows us to fully exploit the sequential nature of our species sampling problem. We empirically show that our approach outperforms state-of-the-art methods and achieves near-Oracle performance on simulated and scRNA-seq data alike. HPY-TS code is available at https://github.com/fedfer/HPYsinglecell.

Unlike standard linear regression, quantile regression captures the relationship between covariates and the conditional response distribution as a whole, rather than only the relationship between covariates and the expected value of the conditional response. However, while there are well-established quantile regression methods for continuous variables and some forms of discrete data, there is no widely accepted method for ordinal variables, despite their importance in many medical contexts. In this work, we describe two existing ordinal quantile regression methods and demonstrate their weaknesses. We then propose a new method, Bayesian ordinal quantile regression with a partially collapsed Gibbs sampler (BORPS). We show superior results using BORPS versus existing methods on an extensive set of simulations. We further illustrate the benefits of our method by applying BORPS to the Fragile Families and Child Wellbeing Study data to tease apart associations with early puberty among both genders. Software is available at: GitHub.com/igrabski/borps.

Characterization of the molecular function of the human genome and its variation across individuals is essential for identifying the cellular mechanisms that underlie human genetic traits and diseases. The Genotype-Tissue Expression (GTEx) project aims to characterize variation in gene expression levels across individuals and diverse tissues of the human body, many of which are not easily accessible. Here we describe genetic effects on gene expression levels across 44 human tissues. We find that local genetic variation affects gene expression levels for the majority of genes, and we further identify inter-chromosomal genetic effects for 93 genes and 112 loci. On the basis of the identified genetic effects, we characterize patterns of tissue specificity, compare local and distal effects, and evaluate the functional properties of the genetic effects. We also demonstrate that multi-tissue, multi-individual data can be used to identify genes and pathways affected by human disease-associated variation, enabling a mechanistic interpretation of gene regulation and the genetic basis of disease.

In the scenario of real-time monitoring of hospital patients, high-quality inference of patients' health status using all information available from clinical covariates and lab tests is essential to enable successful medical interventions and improve patient outcomes. Developing a computational framework that can learn from observational large-scale electronic health records (EHRs) and make accurate real-time predictions is a critical step. In this work, we develop and explore a Bayesian nonparametric model based on Gaussian process (GP) regression for hospital patient monitoring. We propose MedGP, a statistical framework that incorporates 24 clinical and lab covariates and supports a rich reference data set from which relationships between observed covariates may be inferred and exploited for high-quality inference of patient state over time. To do this, we develop a highly structured sparse GP kernel to enable tractable computation over tens of thousands of time points while estimating correlations among clinical covariates, patients, and periodicity in patient observations. MedGP has a number of benefits over current methods, including (i) not requiring an alignment of the time series data, (ii) quantifying confidence regions in the predictions, (iii) exploiting a vast and rich database of patients, and (iv) inferring interpretable relationships among clinical covariates. We evaluate and compare results from MedGP on the task of online prediction for three patient subgroups from two medical data sets across 8,043 patients. We found MedGP improves online prediction over baseline methods for nearly all covariates across different disease subgroups and studies. The publicly available code is at https://github.com/bee-hive/MedGP., Comment: Add new results and appendices

The management of invasive mechanical ventilation, and the regulation of sedation and analgesia during ventilation, constitutes a major part of the care of patients admitted to intensive care units. Both prolonged dependence on mechanical ventilation and premature extubation are associated with increased risk of complications and higher hospital costs, but clinical opinion on the best protocol for weaning patients off of a ventilator varies. This work aims to develop a decision support tool that uses available patient information to predict time-to-extubation readiness and to recommend a personalized regime of sedation dosage and ventilator support. To this end, we use off-policy reinforcement learning algorithms to determine the best action at a given patient state from sub-optimal historical ICU data. We compare treatment policies from fitted Q-iteration with extremely randomized trees and with feedforward neural networks, and demonstrate that the policies learnt show promise in recommending weaning protocols with improved outcomes, in terms of minimizing rates of reintubation and regulating physiological stability.

Expression quantitative trait locus (eQTL) mapping provides a powerful means to identify functional variants influencing gene expression and disease pathogenesis. We report the identification of cis-eQTLs from 7,051 post-mortem samples representing 44 tissues and 449 individuals as part of the Genotype-Tissue Expression (GTEx) project. We find a cis-eQTL for 88% of all annotated protein-coding genes, with one-third having multiple independent effects. We identify numerous tissue-specific cis-eQTLs, highlighting the unique functional impact of regulatory variation in diverse tissues. By integrating large-scale functional genomics data and state-of-the-art fine-mapping algorithms, we identify multiple features predictive of tissue-specific and shared regulatory effects. We improve estimates of cis-eQTL sharing and effect sizes using allele specific expression across tissues. Finally, we demonstrate the utility of this large compendium of cis-eQTLs for understanding the tissue-specific etiology of complex traits, including coronary artery disease. The GTEx project provides an exceptional resource that has improved our understanding of gene regulation across tissues and the role of regulatory variation in human genetic diseases.

Extraversion is a relatively stable and heritable personality trait associated with numerous psychosocial, lifestyle and health outcomes. Despite its substantial heritability, no genetic variants have been detected in previous genome-wide association (GWA) studies, which may be due to relatively small sample sizes of those studies. Here, we report on a large meta-analysis of GWA studies for extraversion in 63,030 subjects in 29 cohorts. Extraversion item data from multiple personality inventories were harmonized across inventories and cohorts. No genome-wide significant associations were found at the single nucleotide polymorphism (SNP) level but there was one significant hit at the gene level for a long non-coding RNA site (LOC101928162). Genome-wide complex trait analysis in two large cohorts showed that the additive variance explained by common SNPs was not significantly different from zero, but polygenic risk scores, weighted using linkage information, significantly predicted extraversion scores in an independent cohort. These results show that extraversion is a highly polygenic personality trait, with an architecture possibly different from other complex human traits, including other personality traits. Future studies are required to further determine which genetic variants, by what modes of gene action, constitute the heritable nature of extraversion.

Domain adaptation addresses the problem created when training data is generated by a so-called source distribution, but test data is generated by a significantly different target distribution. In this work, we present approximate label matching (ALM), a new unsupervised domain adaptation technique that creates and leverages a rough labeling on the test samples, then uses these noisy labels to learn a transformation that aligns the source and target samples. We show that the transformation estimated by ALM has favorable properties compared to transformations estimated by other methods, which do not use any kind of target labeling. Our model is regularized by requiring that a classifier trained to discriminate source from transformed target samples cannot distinguish between the two. We experiment with ALM on simulated and real data, and show that it outperforms techniques commonly used in the field.

Model-based collaborative filtering analyzes user-item interactions to infer latent factors that represent user preferences and item characteristics in order to predict future interactions. Most collaborative filtering algorithms assume that these latent factors are static, although it has been shown that user preferences and item perceptions drift over time. In this paper, we propose a conjugate and numerically stable dynamic matrix factorization (DCPF) based on compound Poisson matrix factorization that models the smoothly drifting latent factors using Gamma-Markov chains. We propose a numerically stable Gamma chain construction, and then present a stochastic variational inference approach to estimate the parameters of our model. We apply our model to time-stamped ratings data sets: Netflix, Yelp, and Last.fm, where DCPF achieves a higher predictive accuracy than state-of-the-art static and dynamic factorization models.

ImportanceNeuroticism is a pervasive risk factor for psychiatric conditions. It genetically overlaps with major depressive disorder (MDD) and is therefore an important phenotype for psychiatric genetics. The Genetics of Personality Consortium has created a resource for genome-wide association analyses of personality traits in more than 63,000 participants (including MDD cases).ObjectivesTo identify genetic variants associated with neuroticism by performing a meta-analysis of genome-wide association results based on 1000 Genomes imputation; to evaluate whether common genetic variants as assessed by single-nucleotide polymorphisms (SNPs) explain variation in neuroticism by estimating SNP-based heritability; and to examine whether SNPs that predict neuroticism also predict MDD.Design, setting, and participantsGenome-wide association meta-analysis of 30 cohorts with genome-wide genotype, personality, and MDD data from the Genetics of Personality Consortium. The study included 63,661 participants from 29 discovery cohorts and 9786 participants from a replication cohort. Participants came from Europe, the United States, or Australia. Analyses were conducted between 2012 and 2014.Main outcomes and measuresNeuroticism scores harmonized across all 29 discovery cohorts by item response theory analysis, and clinical MDD case-control status in 2 of the cohorts.ResultsA genome-wide significant SNP was found on 3p14 in MAGI1 (rs35855737; P = 9.26 × 10-9 in the discovery meta-analysis). This association was not replicated (P = .32), but the SNP was still genome-wide significant in the meta-analysis of all 30 cohorts (P = 2.38 × 10-8). Common genetic variants explain 15% of the variance in neuroticism. Polygenic scores based on the meta-analysis of neuroticism in 27 cohorts significantly predicted neuroticism (1.09 × 10-12

Genome-wide association studies have proven to be essential for understanding the genetic basis of disease. However, many complex traits---personality traits, facial features, disease subtyping---are inherently high-dimensional, impeding simple approaches to association mapping. We developed a nonparametric Bayesian reduced rank regression model for multi-SNP, multi-trait association mapping that does not require the rank of the linear subspace to be specified. We show in simulations and real data that our model shares strength over SNPs and over correlated traits, improving statistical power to identify genetic associations with an interpretable, SNP-supervised low-dimensional linear projection of the high-dimensional phenotype. On the HapMap phase 3 gene expression QTL study data, we identify pleiotropic expression QTLs that classical univariate tests are underpowered to find and that two step approaches cannot recover. Our Python software, BERRRI, is publicly available at GitHub: https://github.com/ashlee1031/BERRRI.

Clustering algorithms start with a fixed divergence, which captures the possibly asymmetric distance between a sample and a centroid. In the mixture model setting, the sample distribution plays the same role. When all attributes have the same topology and dispersion, the data are said to be homogeneous. If the prior knowledge of the distribution is inaccurate or the set of plausible distributions is large, an adaptive approach is essential. The motivation is more compelling for heterogeneous data, where the dispersion or the topology differs among attributes. We propose an adaptive approach to clustering using classes of parametrized Bregman divergences. We first show that the density of a steep exponential dispersion model (EDM) can be represented with a Bregman divergence. We then propose AdaCluster, an expectation-maximization (EM) algorithm to cluster heterogeneous data using classes of steep EDMs. We compare AdaCluster with EM for a Gaussian mixture model on synthetic data and nine UCI data sets. We also propose an adaptive hard clustering algorithm based on Generalized Method of Moments. We compare the hard clustering algorithm with k-means on the UCI data sets. We empirically verified that adaptively learning the underlying topology yields better clustering of heterogeneous data., Comment: Submitted to JMLR on January 2017

Angiotensin-converting enzyme 2 (ACE2) and serine protease TMPRSS2 have been implicated in cell entry for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the virus responsible for coronavirus disease 2019 (COVID-19). The expression of ACE2 and TMPRSS2 in the lung epithelium might have implications for the risk of SARS-CoV-2 infection and severity of COVID-19. We use human genetic variants that proxy angiotensin-converting enzyme (ACE) inhibitor drug effects and cardiovascular risk factors to investigate whether these exposures affect lung ACE2 and TMPRSS2 gene expression and circulating ACE2 levels. We observed no consistent evidence of an association of genetically predicted serum ACE levels with any of our outcomes. There was weak evidence for an association of genetically predicted serum ACE levels with ACE2 gene expression in the Lung eQTL Consortium (p = 0.014), but this finding did not replicate. There was evidence of a positive association of genetic liability to type 2 diabetes mellitus with lung ACE2 gene expression in the Gene-Tissue Expression (GTEx) study (p = 4 × 10−4) and with circulating plasma ACE2 levels in the INTERVAL study (p = 0.03), but not with lung ACE2 expression in the Lung eQTL Consortium study (p = 0.68). There were no associations of genetically proxied liability to the other cardiometabolic traits with any outcome. This study does not provide consistent evidence to support an effect of serum ACE levels (as a proxy for ACE inhibitors) or cardiometabolic risk factors on lung ACE2 and TMPRSS2 expression or plasma ACE2 levels.

Angiotensin-converting enzyme 2 (ACE2) and serine protease TMPRSS2 have been implicated in cell entry for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the virus responsible for coronavirus disease 2019 (COVID-19). The expression of ACE2 and TMPRSS2 in the lung epithelium might have implications for the risk of SARS-CoV-2 infection and severity of COVID-19. We use human genetic variants that proxy angiotensin-converting enzyme (ACE) inhibitor drug effects and cardiovascular risk factors to investigate whether these exposures affect lung ACE2 and TMPRSS2 gene expression and circulating ACE2 levels. We observed no consistent evidence of an association of genetically predicted serum ACE levels with any of our outcomes. There was weak evidence for an association of genetically predicted serum ACE levels with ACE2 gene expression in the Lung eQTL Consortium (p = 0.014), but this finding did not replicate. There was evidence of a positive association of genetic liability to type 2 diabetes mellitus with lung ACE2 gene expression in the Gene-Tissue Expression (GTEx) study (p = 4 × 10−4) and with circulating plasma ACE2 levels in the INTERVAL study (p = 0.03), but not with lung ACE2 expression in the Lung eQTL Consortium study (p = 0.68). There were no associations of genetically proxied liability to the other cardiometabolic traits with any outcome. This study does not provide consistent evidence to support an effect of serum ACE levels (as a proxy for ACE inhibitors) or cardiometabolic risk factors on lung ACE2 and TMPRSS2 expression or plasma ACE2 levels.

Single-cell technologies characterize complex cell populations across multiple data modalities at unprecedented scale and resolution. Multi-omic data for single cell gene expression, in situ hybridization, or single cell chromatin states are increasingly available across diverse tissue types. When isolating specific cell types from a sample of disassociated cells or performing in situ sequencing in collections of heterogeneous cells, one challenging task is to select a small set of informative markers that robustly enable the identification and discrimination of specific cell types or cell states as precisely as possible. Given single cell RNA-seq data and a set of cellular labels to discriminate, scGeneFit selects gene markers that jointly optimize cell label recovery using label-aware compressive classification methods. This results in a substantially more robust and less redundant set of markers than existing methods, most of which identify markers that separate each cell label from the rest. When applied to a data set given a hierarchy of cell types as labels, the markers found by our method improves the recovery of the cell type hierarchy with fewer markers than existing methods using a computationally efficient and principled optimization., The selection of a small set of cellular labels to distinguish a subpopulation of cells from a complex mixture is an important task in cell biology. Here the authors propose a method for supervised genetic marker selection using linear programming and provides a Python package scGeneFit that implements this approach.