Search

The prediction of disease risks is an essential part of personalized medicine, which includes early disease detection, prevention, and intervention. The polygenic risk score (PRS) has become the standard for quantifying genetic liability in predicting disease risks. PRS utilizes single-nucleotide polymorphisms (SNPs) with genetic risks elucidated by genome-wide association studies (GWASs) and is calculated as weighted sum scores of these SNPs with genetic risks using their effect sizes from GWASs as their weights. The utilities of PRS have been explored in many common diseases, such as cancer, coronary artery disease, obesity, and diabetes, and in various non-disease traits, such as clinical biomarkers. These applications demonstrated that PRS could identify a high-risk subgroup of these diseases as a predictive biomarker and provide information on modifiable risk factors driving health outcomes. On the other hand, there are several limitations to implementing PRSs in clinical practice, such as biased sensitivity for the ethnic background of PRS calculation and geographical differences even in the same population groups. Also, it remains unclear which method is the most suitable for the prediction with high accuracy among numerous PRS methods developed so far. Although further improvements of its comprehensiveness and generalizability will be needed for its clinical implementation in the future, PRS will be a powerful tool for therapeutic interventions and lifestyle recommendations in a wide range of diseases. Thus, it may ultimately improve the health of an entire population in the future.

The analysis of RNA-seq gene expression data is often organised aroundtwo successive steps : i) clustering using all of the genes to group the observation units (pa-tients, cells, etc.) into separate and homogeneous subgroups ; then ii) differential analysisof individual genes using hypothesis testing to identify which genes, i.e. which variables,are differentially expressed between the subgroups. However, several subgroups construc-ted in i) can actually contain only units coming from the same homogeneous population :clustering will then artificially create differences between those spurious subgroups, lea-ding to false positives in ii). We propose two inference methods to take into account theinitial clustering step for differential analysis and thus guarantee an effective control of thetype I error. This first method is based on the concept of selective inference while the se-cond one use unimodality and multimodality to describe the separation between clusters.We evaluate the performance of both approaches in extensive numerical simulations aswell as in an application to a real, low dimensional dataset. Both proposed methods leadto valid p-values under their null hypothesis of no difference between subgroups in expres-sion at a selected gene independently of the clustering, while maintaining good statisticalpower. In high dimension, this type I error inflation can be overcome by the dilution of theclustering information, provided that the variables are independent. Yet, in the presenceof correlation (as for gene expression), spurious clusters appear, even though they are notseparable. An adaptation of the above methods to this high dimensional context is thusnecessary., L’analyse des données d’expression génique est souvent organisée autour de deux étapes successives : i) une classification non supervisée utilisant l’ensemble des gènes pour regrouper les unités d’observations (patients, échantillons ou cellules) en sous-groupes distincts et homogènes ; puis ii) l’analyse différentielle se faisant à l’aide de tests d’hypothèse visant à identifier quels gènes, c’est-à-dire quelles variables, sont différentiellement exprimés entre ces sous-groupes. Cependant, cette approche utilisant les même données lors des deux étapes ne permet pas de garantir un bon contrôle de l’erreur de type I à l’étape ii). Nous proposons deux méthodes d’inférence pour tenir compte de l’étape initiale de classification non supervisée lors de l’analyse différentielle et ainsi garantir un contrôle effectif de l’erreur de type I. La première méthode se base sur le concept d’inférence sélective tandis que la seconde repose sur une définition de la séparation de classes faisant uniquement intervenir les concepts d’unimodalité et de multimodalité. Nous avons évalué les performances des deux méthodes grâces à différentes simulations numériques, ainsi que dans une application sur un jeu de données réelles de faible dimension. Les méthodes proposées conduisent à des p-valeurs valides sous l’hypothèse nulle d’absence de différence entre les sous-groupes dans l’expression d’un gène sélectionné, indépendamment de la classification, tout en conservant une bonne puissance statistique. En grande dimension, cette inflation de l’erreur de type I peut-être contre-balancée par la dilution du signal utilisé pour la classification, à condition que les variables soient indépendantes. En revanche, en présence de corrélation (comme c’est le cas en pratique pour l’expression génique), des classes artificielles apparaissent alors que celles-ci ne sont pas séparables.Une adaptation des méthodes à ce contexte de grande dimension est donc nécessaire.

Statistical methods for genomewide association studies (GWAS) continue to improve. However, the increasing volume and variety of genetic and genomic data make computational speed and ease of data manipulation mandatory in future software. In our view, a collaborative effort of statistical geneticists is required to develop open source software targeted to genetic epidemiology. Our attempt to meet this need is called the OPENMENDELproject (https://openmendel.github.io). It aims to (1) enable interactive and reproducible analyses with informative intermediate results, (2) scale to big data analytics, (3) embrace parallel and distributed computing, (4) adapt to rapid hardware evolution, (5) allow cloud computing, (6) allow integration of varied genetic data types, and (7) foster easy communication between clinicians, geneticists, statisticians, and computer scientists. This article reviews and makes recommendations to the genetic epidemiology community in the context of the OPENMENDEL project., 16 pages, 2 figures, 2 tables

This tutorial is a learning resource that outlines the basic process and provides specific software tools for implementing a complete genome‐wide association analysis. Approaches to post‐analytic visualization and interrogation of potentially novel findings are also presented. Applications are illustrated using the free and open‐source R statistical computing and graphics software environment, Bioconductor software for bioinformatics and the UCSC Genome Browser. Complete genome‐wide association data on 1401 individuals across 861,473 typed single nucleotide polymorphisms from the PennCATH study of coronary artery disease are used for illustration. All data and code, as well as additional instructional resources, are publicly available through the Open Resources in Statistical Genomics project: http://www.stat-gen.org. © 2015 The Authors. Statistics in Medicine Published by John Wiley & Sons Ltd.

Background: Recent large-scale undertakings such as ENCODE and Roadmap Epigenomics have generated experimental data mapped to the human reference genome (as genomic tracks) representing a variety of functional elements across a large number of cell types. Despite the high potential value of these publicly available data for a broad variety of investigations, little attention has been given to the analytical methodology necessary for their widespread utilisation. Findings: We here present a first principled treatment of the analysis of collections of genomic tracks. We have developed novel computational and statistical methodology to permit comparative and confirmatory analyses across multiple and disparate data sources. We delineate a set of generic questions that are useful across a broad range of investigations and discuss the implications of choosing different statistical measures and null models. Examples include contrasting analyses across different tissues or diseases. The methodology has been implemented in a comprehensive open-source software system, the GSuite HyperBrowser. To make the functionality accessible to biologists, and to facilitate reproducible analysis, we have also developed a web-based interface providing an expertly guided and customizable way of utilizing the methodology. With this system, many novel biological questions can flexibly be posed and rapidly answered. Conclusions: Through a combination of streamlined data acquisition, interoperable representation of dataset collections, and customizable statistical analysis with guided setup and interpretation, the GSuite HyperBrowser represents a first comprehensive solution for integrative analysis of track collections across the genome and epigenome. The software is available at: https://hyperbrowser.uio.no., This work was supported by the Research Council of Norway (under grant agreements 221580, 218241, and 231217/F20), by the Norwegian Cancer Society (under grant agreements 71220’PR-2006-0433 and 3485238-2013), and by the South-Eastern Norway Regional Health Authority (under grant agreement 2014041).

For several decades, sequence alignment has been a widely used tool in bioinformatics. For instance, finding homologous sequences with a known function in large databases is used to get insight into the function of nonannotated genomic regions. Very efficient tools like BLAST have been developed to identify and rank possible homologous sequences. To estimate the significance of the homology, the ranking of alignment scores takes a background model for random sequences into account. Using this model we can estimate the probability to find two exactly matching subsequences by chance in two unrelated sequences. For two homologous sequences, the corresponding probability is much higher, which allows us to identify them. Here we focus on the distribution of lengths of exact sequence matches between protein-coding regions of pairs of evolutionarily distant genomes. We show that this distribution exhibits a power-law tail with an exponent α=−5. Developing a simple model of sequence evolution by substitutions and segmental duplications, we show analytically and computationally that paralogous and orthologous gene pairs contribute differently to this distribution. Our model explains the differences observed in the comparison of coding and noncoding parts of genomes, thus providing a better understanding of statistical properties of genomic sequences and their evolution.

Some important and interesting topics in the newly emerging disciplines of Statistical genomics andbioinformatics have been discussed briefly in relation to plants with possible references to fruit crops. This paper is therefore divided into two parts relating to the two disciplines, respectively. In the first part, mapping of quantitative trait loci (QTL), association mapping, mapping of gene expression transcripts (eQTL), marker-assisted selection, and a systems approach to quantitative genetics have been dealt with. In the second part, generation of databases, annotation, annotated sequence databases, and sequence similarity search have been described.

Background Oncology is a field that profits tremendously from the genomic data generated by high-throughput technologies, including next-generation sequencing. However, in order to exploit, integrate, visualize and interpret such high-dimensional data efficiently, non-trivial computational and statistical analysis methods are required that need to be developed in a problem-directed manner. Discussion For this reason, computational cancer biology aims to fill this gap. Unfortunately, computational cancer biology is not yet fully recognized as a coequal field in oncology, leading to a delay in its maturation and, as an immediate consequence, an under-exploration of high-throughput data for translational research. Summary Here we argue that this imbalance, favoring ’wet lab-based activities’, will be naturally rectified over time, if the next generation of scientists receives an academic education that provides a fair and competent introduction to computational biology and its manifold capabilities. Furthermore, we discuss a number of local educational provisions that can be implemented on university level to help in facilitating the process of harmonization.

Integrating biological knowledge in gene expression data analysis

Conselho Nacional de Desenvolvimento Científico e Tecnológico Como os recentes avanços na tecnologia têm levado ao desenvolvimento de novas tecnologias de genotipagem, no futuro, é mais provável que os mapas de ligação de alta densidade serão construídos a partir de marcadores dominantes e co-dominantes. Recentemente, uma abordagem estritamente genética foi proposta para a estimação da freqüência de recombinação (r) entre marcadores co-dominantes em famílias de irmãos completos. O conjunto completo de estimadores quase foi obtido, mas infelizmente, uma configuração envolvendo a estimativa da distância entre os marcadores dominantes, que segregam na proporção 3:1 e marcadores co-dominantes, não foi levada em consideração. Aqui novos nove estimadores são acrescentados ao conjunto previamente publicado, tornando possível cobrir todas as combinações de marcadores moleculares com dois a quatro alelos (sem epistasia) em uma família de irmãos completos. Isso inclui a segregação em um ou ambos os genitores, dominância e todas as configurações de fases de ligação. Como populações de retrocruzamentos (RC) são frequentemente utilizadas como populações de mapeamento, tanto em espécies autógamas, quanto em espécies alógamas foi conduzido um estudo de simulação para testar as implicações do tamanho da população, herdabilidade da característica, propriedades do QTL (r2, a e posição) e densidade de marcadores no poder de detecção e precisão do mapeamento de QTLs. Para tanto foram simuladas populações com diferentes tamanhos, com diferentes características (h2, número de QTLs e posição) e os dados analisados com dois métodos de mapeamento de QTLs comumente utilizados (mapeamento por intervalo simples (MIS) e mapeamento por intervalo composto (MIC)). Verificou-se que o tamanho da amostra tem uma grande implicação no poder de detecção e como conseqüência na estimação da magnitude da variação explicada pelo QTL e no efeito genético, em função de populações pequenas não permitirem o mapeamento de QTLs de pequeno efeito, principalmente quando esses estão envolvidos no controle genético de características de baixa herdabilidade. Também foi verificado que o posicionamento de QTLs baseados em MIC é mais acurado que MIS e que em média os QTLs mapeados estavam próximos as suas posições simuladas. Um resultado interessante é que o MIC tende a subestimar os valores de magnitude (r2) especialmente em populações grandes/ características de baixa herdabilidade e superestimá-la em populações pequenas, o que pode ser um reflexo do pequeno coeficiente de variação do erro utilizado, ou devido ao fato de quando os marcadores não se encontram na exata posição do QTL, esse parâmetro é de fato esperado ser subestimado. Destaca-se também, o fato que quando marcadores estão amplamente distribuídos ao longo do genoma (~10cM), e desse modo cobrindo a região do QTL, se um dos marcadores já estiver próximo ao QTL, um maior número de marcadores (~1cM) não melhora a precisão do mapeamento do QTL em populações suficientemente grandes. Baseado nesses resultados recomenda-se o uso de populações de tamanho adequado, ≥500, se a intenção é mapear QTLs em populações de RC, porque nessa situação, mesmo mapas de média densidade podem ser usados para mapear QTLs de grande ou pequeno efeito com grande confiabilidade. Finalmente, como os procedimentos de mapeamento de ligação e mapeamento de QTLs em famílias de irmãos completos (FIC) de espécies alógamas são bastante diversos, foi conduzido um estudo comparando o método de mapeamento por pseudo-testcross modificado (PST) (usando microsatélites), com o método de mapeamento baseado na FIC; em termos de ordenamento dos marcadores, distância entre os marcadores, comprimento total do mapa, variância das estimativas de distância e estresse. Investigou-se também o poder de detecção e a precisão de métodos de mapeamento de QTLs por intervalos baseados nos mapas PST ou no mapa para a FIC. Verificou-se que em geral as duas estratégias geram mapas altamente correlacionados com comprimentos dos grupos de ligação proporcionais. Verificou-se também que independentemente da abordagem de mapeamento de QTLs utilizadas, o poder de detecção é reduzido em populações pequenas, especialmente em situações onde a herdabilidade da característica ou magnitude do QTL é pequena. Também foi verificado que apesar dos dois métodos serem aparentemente equivalentes em termos de posicionamento do QTL para características de alta herdabilidade/ QTLs de grande efeito, o MIC baseado nos mapas pseudo-testcross prove dados mais acurados para características de baixa herdabilidade/QTLs de pequeno efeito. Como relação à magnitude dos QTLs, notou-se que ambos os métodos parecem ser equivalentes, sendo os valores superestimados para características de alta herdabilidade e subestimados para características de baixa herdabilidade, independentemente do tamanho amostral. Assim para espécies alógamas com médio nível de recursos genômicos, propõem-se que tanto a abordagem de PST quanto a abordagem baseada na FIC, e métodos de mapeamento de QTLs relacionados, possam ser utilizados para gerar mapas genéticos e mapear QTLs com alta confiabilidade. É importante ressaltar, entretanto, que outros estudos, usando diferentes cenários, i.e. diferentes coeficientes de variação do erro, diferentes números de QTLs, diferentes distribuições de marcadores, que coletivamente podem tornar a simulação um pouco mais realística, são necessários para verificar que os resultados deste trabalho se mantêm em todas as situações. As high-throughput genomic tools have led to the development of novel genotyping procedures, it is likely that, in the future, high density linkage maps will be constructed from both dominant and co-dominant markers. Recently, a strictly genetic approach was described for estimating the recombination frequency (r) between co-dominant markers in full-sib families. The complete set of maximum likelihood estimators for r in full-sib families was almost obtained, but unfortunately, one particular configuration involving dominant markers, segregating in a 3:1 ratio and co-dominant markers, was not considered. Here we add nine further estimators to the previously published set, thereby making it possible to cover all combinations of molecular markers with two to four alleles (without epistasis) in a full-sib family. This includes segregation in one or both parents, dominance and all linkage phase configurations. As backcross (BC) populations are often used as mapping populations both in self pollinating species, and in out-breeding species we also undertook a simulation study to test implications of population size, trait heritability, QTL properties (r2, a and position) and marker density in the power and precision of QTL mapping. For that we have simulated populations with different sizes, with different characteristics (h2, QTL number and location) and the data analyzed with two xv QTL mapping methods (simple interval mapping (SIM) and composite interval mapping (CIM)). We found that sample size has a major implication in the detection power and as consequence in the estimation of the magnitude and additive genetic effect, as small populations do not allow mapping of low effect QTLs, especially if these QTLs are involved in the genetic control of traits with low heritability. We also found that the positioning of the QTLs based on CIM is more accurate than SIM and that on average the mapped QTLs are close to their simulated position. The results showed that CIM tend to underestimate the magnitude (r2) values especially in large population sizes/low heritabilities traits and overestimate it in smaller populations, which can be a reflection of the low coefficient of variation of the error used, or due to fact that when markers aren´t in the same of the QTL, this parameter is indeed expected to be underestimated. We also highlight the fact, that when markers are evenly distributed across the genome (~10 cM), and therefore covering the QTL region, if one of the markers is already close to the QTL, larger number of markers (~ 1cM) do not improve the precision of QTL mapping in sufficiently large mapping populations. Based on our results we recommend the use of adequate sample size, say ≥500, if the intention is map QTLs in backcross populations, because in this situation even mid-density genetic maps can be used to map QTLs of large or small effect with high confidence. Finally, as the procedures for linkage and QTL mapping in fullsib families of outbreeding species are quite diverse, we also undertook a simulation study comparing the modified pseudo-testcross (using SSR markers) and the full-sib mapping designs in terms of marker ordering, distance between markers, total map size, distance variance and stress. We also investigated the power and precision of interval mapping procedures based on the full-sib and on the modified pseudo-testcross maps. We show that in general the modified pseudo-testcross and the full-sib mapping designs generate highly correlated maps with proportional linkage groups length. That independent of the QTL mapping approach used, detection power is reduced in small populations, especially in situations where trait heritability or QTL magnitude are low. We also found that although both methods appear to be equivalent in terms of QTL positioning for high heritability traits/major effect QTLs, the CIM based on modified pseudo-testcross maps provide more accurate data for low heritability traits/minor effect QTLs in larger populations. With regard to QTLs magnitude, we show that both methods appear to be equivalent, and that the magnitude values tended to be over estimated for the high heritability trait, and underestimated for the low heritability trait, independent of the sample size. Thus, for outbreeding species with mid-level of genomic resources we propose that either the modified pseudo-testcross or the single full-sib mapping design and the related QTL mapping strategies can be used to generate genetic maps and map QTLs with high confidence. It is important to highlight however, that, other studies, using different scenarios, i.e. different coefficients of variation of the error, different number of QTLs, different marker distributions, which collectively may make the simulation a bit more realistic, are needed in order to see if the results of our work hold true in every situation.

As high-throughput genomic tools, such as the DNA microarray platform, have lead to the development of novel genotyping procedures, such as Diversity Arrays Technology (DArT) and Single Nucleotide Polymorphisms (SNPs), it is likely that, in the future, high density linkage maps will be constructed from both dominant and co-dominant markers. Recently, a strictly genetic approach was described for estimating recombination frequency (r) between co-dominant markers in full-sib families. The complete set of maximum likelihood estimators for r in full-sib families was almost obtained, but unfortunately, one particular configuration involving dominant markers, segregating in a 3:1 ratio and co-dominant markers, was not considered. Here we add nine further estimators to the previously published set, thereby making it possible to cover all combinations of molecular markers with two to four alleles (without epistasis) in a full-sib family. This includes segregation in one or both parents, dominance and all linkage phase configurations.

While multiple testing procedures have been the focus of much statistical research, an important facet of the problem is how to deal with possible confounding. Procedures have been developed by authors in genetics and statistics. In this chapter, we relate these proposals. We propose two new multiple testing approaches within this framework. The first combines sensitivity analysis methods with false discovery rate estimation procedures. The second involves construction of shrinkage estimators that utilize the mixture model for multiple testing. The procedures are illustrated with applications to a gene expression profiling experiment in prostate cancer., Comment: Published in at http://dx.doi.org/10.1214/193940307000000176 the IMS Collections (http://www.imstat.org/publications/imscollections.htm) by the Institute of Mathematical Statistics (http://www.imstat.org)