Search

Visual Question Answering (VQA) models often perform poorly onout-of-distribution data and struggle on domain generalization. Due to themulti-modal nature of this task, multiple factors of variation are intertwined,making generalization difficult to analyze. This motivates us to introduce avirtual benchmark, Super-CLEVR, where different factors in VQA domain shiftscan be isolated in order that their effects can be studied independently. Fourfactors are considered: visual complexity, question redundancy, conceptdistribution and concept compositionality. With controllably generated data,Super-CLEVR enables us to test VQA methods in situations where the test datadiffers from the training data along each of these axes. We study four existingmethods, including two neural symbolic methods NSCL and NSVQA, and twonon-symbolic methods FiLM and mDETR; and our proposed method, probabilisticNSVQA (P-NSVQA), which extends NSVQA with uncertainty reasoning. P-NSVQAoutperforms other methods on three of the four domain shift factors. Ourresults suggest that disentangling reasoning and perception, combined withprobabilistic uncertainty, form a strong VQA model that is more robust todomain shifts. The dataset and code are released athttps://github.com/Lizw14/Super-CLEVR.

Genomic global positioning system (GPS) applies the multilateration technique commonly used in the GPS to genomic data. In the framework we present here, investigators calculate genetic distances from their samples to reference samples, which are from data held in the public domain, and share this information with others. This sharing enables certain types of genomic analysis, such as identifying sample overlaps and close relatives, decomposing ancestry, and mapping of geographical origin without disclosing personal genomes. Thus, our method can be seen as a balance between open data sharing and privacy protection. Electronic supplementary material The online version of this article (10.1186/s13059-019-1792-2) contains supplementary material, which is available to authorized users.

The Genotype-Tissue Expression (GTEx) project dissects how genetic variation affects gene expression and splicing. Some human genetic variants affect the amount of RNA produced and the splicing of gene transcripts, crucial steps in development and maintaining a healthy individual. However, some of these changes only occur in a small number of tissues within the body. The Genotype-Tissue Expression (GTEx) project has been expanded over time, and, looking at the final data in version 8, Aguet et al. present a deep characterization of genetic associations and gene expression and splicing in 838 individuals over 49 tissues (see the Perspective by Wilson). This large study was able to characterize the details underlying many aspects of gene expression and provides a resource with which to better understand the fundamental molecular mechanisms of how genetic variants affect gene regulation and complex traits in humans. Science, this issue p. 1318; see also p. 1298 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 version 8 data, examining 15,201 RNA-sequencing samples from 49 tissues of 838 postmortem donors. We comprehensively characterize genetic associations for gene expression and splicing in cis and trans, 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. We thank the donors and their families for their generous gifts of organ donation for transplantation and tissue donations for the GTEx research project; the Genomics Platform at the Broad Institute for data generation; J. Struewing for support and leadership of the GTEx project; M. Khan and C. Stolte for the illustrations in Fig. 1; and R. Do, D. Jordan, and M. Verbanck for providing GWAS pleiotropy scores. Funding: This work was supported by the Common Fund of the Office of the Director, U.S. National Institutes of Health, and by NCI, NHGRI, NHLBI, NIDA, NIMH, NIA, NIAID, and NINDS through NIH contracts HHSN261200800001E (Leidos Prime contract with NCI: A.M.S., D.E.T., N.V.R., J.A.M., L.S., M.E.B., L.Q., T.K., D.B., K.R., and A.U.), 10XS170 (NDRI: W.F.L., J.A.T., G.K., A.M., S.S., R.H., G.Wa., M.J., M.Wa., L.E.B., C.J., J.W., B.R., M.Hu., K.M., L.A.S., H.M.G., M.Mo., and L.K.B.), 10XS171 (Roswell Park Cancer Institute: B.A.F., M.T.M., E.K., B.M.G., K.D.R., and J.B.), 10X172 (Science Care Inc.), 12ST1039 (IDOX), 10ST1035 (Van Andel Institute: S.D.J., D.C.R., and D.R.V.), HHSN268201000029C (Broad Institute: F.A., G.G., K.G.A., A.V.S., X.Li., E.T., S.G., A.G., S.A., K.H.H., D.T.N., K.H., S.R.M., and J.L.N.), 5U41HG009494 (F.A., G.G., and K.G.A.), and through NIH grants R01 DA006227-17 (University of Miami Brain Bank: D.C.M. and D.A.D.), Supplement to University of Miami grant DA006227 (D.C.M. and D.A.D.), R01 MH090941 (University of Geneva), R01 MH090951 and R01 MH090937 (University of Chicago), R01 MH090936 (University of North Carolina–Chapel Hill), R01MH101814 (M.M.-A., V.W., S.B.M., R.G., E.T.D., D.G.-M., and A.V.), U01HG007593 (S.B.M.), R01MH101822 (C.D.B.), U01HG007598 (M.O. and B.E.S.), U01MH104393 (A.P.F.), extension H002371 to 5U41HG002371 (W.J.K.), as well as other funding sources: R01MH106842 (T.L., P.M., E.F., and P.J.H.), R01HL142028 (T.L., Si.Ka., and P.J.H.), R01GM122924 (T.L. and S.E.C.), R01MH107666 (H.K.I.), P30DK020595 (H.K.I.), UM1HG008901 (T.L.), R01GM124486 (T.L.), R01HG010067 (Y.Pa.), R01HG002585 (G.Wa. and M.St.), Gordon and Betty Moore Foundation GBMF 4559 (G.Wa. and M.St.), 1K99HG009916-01 (S.E.C.), R01HG006855 (Se.Ka. and R.E.H.), BIO2015-70777-P, Ministerio de Economia y Competitividad and FEDER funds (M.M.-A., V.W., R.G., and D.G.-M.), la Caixa Foundation ID 100010434 under agreement LCF/BQ/SO15/52260001 (D.G.-M.), NIH CTSA grant UL1TR002550-01 (P.M.), Marie-Skłodowska Curie fellowship H2020 Grant 706636 (S.K.-H.), R35HG010718 (E.R.G.), FPU15/03635, Ministerio de Educación, Cultura y Deporte (M.M.-A.),R01MH109905, 1R01HG010480 (A.Ba.), Searle Scholar Program (A.Ba.), R01HG008150 (S.B.M.), 5T32HG000044-22, NHGRI Institutional Training Grant in Genome Science (N.R.G.), EU IMI program (UE7-DIRECT-115317-1) (E.T.D. and A.V.), FNS funded project RNA1 (31003A_149984) (E.T.D. and A.V.), DK110919 (F.H.), F32HG009987 (F.H.), Massachusetts Lions Eye Research Fund Grant (A.R.H.), Wellcome grant WT108749/Z/15/Z (P.F.), and European Molecular Biology Laboratory (P.F. and D.Z.). Peer Reviewed "Article signat per 1 autors/es del BSC membres del THE GTEX CONSORTIUM: Marta Mele Messeguer"

Advances in whole genome sequencing promise to enable the accurate and comprehensive structural variant (SV) discovery. Dissecting SVs from whole genome sequencing (WGS) data presents a substantial number of challenges and a plethora of SV-detection methods have been developed. Currently, there is a paucity of evidence which investigators can use to select appropriate SV-detection tools. In this paper, we evaluated the performance of SV-detection tools using a comprehensive PCR-confirmed gold standard set of SVs. In contrast to the previous benchmarking studies, our gold standard dataset included a complete set of SVs allowing us to report both precision and sensitivity rates of SV-detection methods. Our study investigates the ability of the methods to detect deletions, thus providing an optimistic estimate of SV detection performance, as the SV-detection methods that fail to detect deletions are likely to miss more complex SVs. We found that SV-detection tools varied widely in their performance, with several methods providing a good balance between sensitivity and precision. Additionally, we have determined the SV callers best suited for low and ultra-low pass sequencing data.

Many disease risk loci identified in genome-wide association studies are present in non-coding regions of the genome. It is hypothesized that these variants affect complex traits by acting as expression quantitative trait loci (eQTLs) that influence expression of nearby genes. This indicates that many causal variants for complex traits are likely to be causal variants for gene expression. Hence, identifying causal variants for gene expression is important for elucidating the genetic basis of not only gene expression but also complex traits. However, detecting causal variants is challenging due to complex genetic correlation among variants known as linkage disequilibrium (LD) and the presence of multiple causal variants within a locus. Although several fine-mapping approaches have been developed to overcome these challenges, they may produce large sets of putative causal variants when true causal variants are in high LD with many non-causal variants. In eQTL studies, there is an additional source of information that can be used to improve fine-mapping called allele-specific expression (ASE) that measures imbalance in gene expression due to different alleles. In this work, we develop a novel statistical method that leverages both ASE and eQTL information to detect causal variants that regulate gene expression. We illustrate through simulations and application to the Genotype-Tissue Expression (GTEx) dataset that our method identifies the true causal variants with higher specificity than an approach that uses only eQTL information. In the GTEx dataset, our method achieves the median reduction rate of 11% in the number of putative causal variants.ContactJaeHoonSul@mednet.ucla.edu, eeskin@cs.ucla.edu

Short Structured AbstractSummaryEvery sequencing library contains duplicate reads. While many duplicates arise during polymerase chain reaction (PCR), some duplicates derive from multiple identical fragments of mRNA present in the original lysate (termed “biological duplicates”). Unique Molecular Identifiers (UMIs) are random oligonucleotide sequences that allow differentiation between technical and biological duplicates. Here we report the development of UMI-Reducer, a new computational tool for processing and differentiating PCR duplicates from biological duplicates. UMI-Reducer uses UMIs and the mapping position of the read to identify and collapse reads that are technical duplicates. Remaining true biological reads are further used for bias-free estimate of mRNA abundance in the original lysate. This strategy is of particular use for libraries made from low amounts of starting material, which typically require additional cycles of PCR and therefore are most prone to PCR duplicate bias.Availability and ImplementationThe UMI-Reducer is an open source Python software and is freely available for non-commercial use (GPL-3.0) at https://sergheimangul.wordpress.com/umi-reducer/. Documentation and tutorials are available at https://github.com/smangul1/UMI-Reducer/wiki/.Contactsmangul@ucla.edu, SVanDriesche@mednet.ucla.eduSupplementary informationFlowchart of Library Construction

Studying genetic determinants of intermediate phenotypes is a powerful tool to increase our understanding of genotype-phenotype correlations. Metabolic traits pertinent to the central nervous system (CNS) constitute a potentially informative target for genetic studies of intermediate phenotypes as their genetic underpinnings may elucidate etiological mechanisms. We therefore conducted a genome-wide association study (GWAS) of monoamine metabolite (MM) levels in cerebrospinal fluid (CSF) of 414 human subjects from the general population. In a linear model correcting for covariates, we identified one locus associated with MMs at a genome-wide significant level (standardized β=0.32, P=4.92 × 10(-8)), located 20 kb from SSTR1, a gene involved with brain signal transduction and glutamate receptor signaling. By subsequent whole-genome expression quantitative trait locus (eQTL) analysis, we provide evidence that this variant controls expression of PDE9A (β=0.21; P unadjusted=5.6 × 10(-7); P corrected=0.014), a gene previously implicated in monoaminergic transmission, major depressive disorder and antidepressant response. A post hoc analysis of loci significantly associated with psychiatric disorders suggested that genetic variation at CSMD1, a schizophrenia susceptibility locus, plays a role in the ratio between dopamine and serotonin metabolites in CSF. The presented DNA and mRNA analyses yielded genome-wide and suggestive associations in biologically plausible genes, two of which encode proteins involved with glutamate receptor functionality. These findings will hopefully contribute to an exploration of the functional impact of the highlighted genes on monoaminergic transmission and neuropsychiatric phenotypes.

Tourette's syndrome (TS) is a developmental disorder that has one of the highest familial recurrence rates among neuropsychiatric diseases with complex inheritance. However, the identification of definitive TS susceptibility genes remains elusive. Here, we report the first genome-wide association study (GWAS) of TS in 1285 cases and 4964 ancestry-matched controls of European ancestry, including two European-derived population isolates, Ashkenazi Jews from North America and Israel and French Canadians from Quebec, Canada. In a primary meta-analysis of GWAS data from these European ancestry samples, no markers achieved a genome-wide threshold of significance (P

Malaria transmission is dependent on the propensity of Anopheles mosquitoes to bite\ud humans (anthropophily) instead of other dead end hosts. Recent increases in the usage of\ud Long Lasting Insecticide Treated Nets (LLINs) in Africa have been associated with reductions in highly anthropophilic and endophilic vectors such as Anopheles gambiae s.s., leaving species with a broader host range, such as Anopheles arabiensis, as the most\ud prominent remaining source of transmission in many settings. An. arabiensis appears to be\ud more of a generalist in terms of its host choice and resting behavior, which may be due to\ud phenotypic plasticity and/or segregating allelic variation. To investigate the genetic basis of host choice and resting behavior in An. arabiensis we sequenced the genomes of 23\ud human-fed and 25 cattle-fed mosquitoes collected both in-doors and out-doors in the Kilombero Valley, Tanzania. We identified a total of 4,820,851 SNPs, which were used to conduct the first genome-wide estimates of “SNP heritability”for host choice and resting\ud behavior in this species. A genetic component was detected for host choice (human vs cow\ud fed; permuted P = 0.002), but there was no evidence of a genetic component for resting\ud behavior (indoors versus outside; permuted P = 0.465). A principal component analysis\ud (PCA) segregated individuals based on genomic variation into three groups which were\ud characterized by differences at the 2Rb and/or 3Ra paracentromeric chromosome inversions. There was a non-random distribution of cattle-fed mosquitoes between the PCA clusters, suggesting that alleles linked to the 2Rb and/or 3Ra inversions may influence host\ud choice. Using a novel inversion genotyping assay, we detected a significant enrichment of\ud the standard arrangement (non-inverted) of 3Ra among cattle-fed mosquitoes (N = 129)\ud versus all non-cattle-fed individuals. Thus, tracking the frequency of the 3Ra in An. arabiensis populations may be of use to infer selection on host choice behavior within these vector populations; possibly in response to vector control. Controlled\ud host-choice assays are needed to discern whether the observed genetic component has a\ud direct relationship with innate host preference. A better understanding of the genetic basis\ud for host feeding behavior in An. arabiensis may also open avenues for novel vector control\ud strategies based on driving genes for zoophily into wild mosquito populations.

Understanding the genetics of gene regulation provides information on the cellular mechanisms through which genetic variation influences complex traits. Expression quantitative trait loci, or eQTLs, are enriched for polymorphisms that have been found to be associated with disease risk. While most analyses of human data has focused on regulation of expression by nearby variants (cis-eQTLs), distal or trans-eQTLs may have broader effects on the transcriptome and important phenotypic consequences, necessitating a comprehensive study of the effects of genetic variants on distal gene transcription levels. In this work, we identify trans-eQTLs in the Genotype Tissue Expression (GTEx) project data1, consisting of 449 individuals with RNA-sequencing data across 44 tissue types. We find 81 genes with a trans-eQTL in at least one tissue, and we demonstrate that trans-eQTLs are more likely than cis-eQTLs to have effects specific to a single tissue. We evaluate the genomic and functional properties of trans-eQTL variants, identifying strong enrichment in enhancer elements and Piwi-interacting RNA clusters. Finally, we describe three tissue-specific regulatory loci underlying relevant disease associations: 9q22 in thyroid that has a role in thyroid cancer, 5q31 in skeletal muscle, and a previously reported master regulator near KLF14 in adipose. These analyses provide a comprehensive characterization of trans-eQTLs across human tissues, which contribute to an improved understanding of the tissue-specific cellular mechanisms of regulatory genetic variation.

The N-methyl-d-aspartate receptor (NMDAR) coagonists glycine, d-serine and l-proline play crucial roles in NMDAR-dependent neurotransmission and are associated with a range of neuropsychiatric disorders. We conducted the first genome-wide association study of concentrations of these coagonists and their enantiomers in plasma and cerebrospinal fluid (CSF) of human subjects from the general population (N=414). Genetic variants at chromosome 22q11.2, located in and near PRODH (proline dehydrogenase), were associated with l-proline in plasma (β=0.29; P=6.38 × 10-10). The missense variant rs17279437 in the proline transporter SLC6A20 was associated with l-proline in CSF (β=0.28; P=9.68 × 10-9). Suggestive evidence of association was found for the d-serine plasma-CSF ratio at the d-amino-acid oxidase (DAO) gene (β=−0.28; P=9.08 × 10-8), whereas a variant in SRR (that encodes serine racemase and is associated with schizophrenia) constituted the most strongly associated locus for the l-serine to d-serine ratio in CSF. All these genes are highly expressed in rodent meninges and choroid plexus, anatomical regions relevant to CSF physiology. The enzymes and transporters they encode may be targeted to further construe the nature of NMDAR coagonist involvement in NMDAR gating. Furthermore, the highlighted genetic variants may be followed up in clinical populations, for example, schizophrenia and 22q11 deletion syndrome. Overall, this targeted metabolomics approach furthers the understanding of NMDAR coagonist concentration variability and sets the stage for non-targeted CSF metabolomics projects.Molecular Psychiatry advance online publication, 10 February 2015; doi:10.1038/mp.2014.190.

Genome structure, variation and function; We report an analysis of allele specific expression [ASE] and parent-of origin expression in adult mouse liver using next generation sequencing (RNA-Seq) of reciprocal crosses of heterozygous F1 mice from the parental strains C57BL/6J and DBA/2J. The genes exhibiting ASE differed markedly from the putative cis-acting expression quantitative trait loci (cis-eQTL) identified in an intercross between the same strains. While about 60% of the ASE, mapped by RNA-Seq, were found in the eQTL gene set, only a small fraction of the eQTL, mapped by linkage analysis, were found in the ASE gene set.We discuss the various biological and technical factors that contribute to these differences, in particular strengths of the two approaches in making a distinction between local and cis eQTL. We also identify genes exhibiting parental imprinting and complex expression patterns. Our study demonstrates the importance of biological replicates, which is not currently the norm, to limit the number of false positives with such RNA-Seq data.

We report an analysis of allele specific expression [ASE] and parent-of origin expression in adult mouse liver using next generation sequencing (RNA-Seq) of reciprocal crosses of heterozygous F1 mice from the parental strains C57BL/6J and DBA/2J. The genes exhibiting ASE differed markedly from the putative cis-acting expression quantitative trait loci (cis-eQTL) identified in an intercross between the same strains. While about 60% of the ASE, mapped by RNA-Seq, were found in the eQTL gene set, only a small fraction of the eQTL, mapped by linkage analysis, were found in the ASE gene set.We discuss the various biological and technical factors that contribute to these differences, in particular strengths of the two approaches in making a distinction between local and cis eQTL. We also identify genes exhibiting parental imprinting and complex expression patterns. Our study demonstrates the importance of biological replicates, which is not currently the norm, to limit the number of false positives with such RNA-Seq data.

We introduce a class of string kernels, called mismatch kernels, for use with support vector machines (SVMs) in a discriminative approach to the protein classification problem. These kernels measure sequence similarity based on shared occurrences of k-length subsequences, counted with up to m mismatches, and do not rely on any generative model for the positive training sequences. We compute the kernels efficiently using a mismatch tree data structure and report experiments on a benchmark SCOP dataset, where we show that the mismatch kernel used with an SVM classifier performs as well as the Fisher kernel, the most successful method for remote homology detection, while achieving considerable computational savings.

There have been many efforts to identify causal graphical features such as directed edges between random variables from observational data. Recently, Tian et al. proposed a new dynamic programming algorithm which computes marginalized posterior probabilities of directed edge features over all the possible structures in O(n3n) time when the number of parents per node is bounded by a constant, where n is the number of variables of interest. However the main drawback of this approach is that deciding a single appropriate threshold for the existence of the directed edge feature is difficult due to the scale difference of the posterior probabilities between the directed edges forming v-structures and the directed edges not forming v-structures. We claim that computing posterior probabilities of both adjacencies and v-structures is necessary and more effective for discovering causal graphical features, since it allows us to find a single appropriate decision threshold for the existence of the feature that we are testing. For efficient computation, we provide a novel dynamic programming algorithm which computes the posterior probabilities of all of n(n – 1)/2 adjacency and n(n–1 choose 2) v-structure features in O(n3 * 3n) time.