Search

The Human Reference Genome serves as the foundation for modern genomic analyses. However, in its present form, it does not adequately represent the vast genetic diversity of the human population. In this study, we explored the consensus genome as a potential successor of the current reference genome and assessed its effect on the accuracy of RNA-seq read alignment. To find the best haploid genome representation, we constructed consensus genomes at the pan-human, superpopulation, and population levels, using variant information from The 1000 Genomes Project Consortium. Using personal haploid genomes as the ground truth, we compared mapping errors for real RNA-seq reads aligned to the consensus genomes versus the reference genome. For reads overlapping homozygous variants, we found that the mapping error decreased by a factor of approximately two to three when the reference was replaced with the pan-human consensus genome. We also found that using more population-specific consensuses resulted in little to no increase over using the pan-human consensus, suggesting a limit in the utility of incorporating a more specific genomic variation. Replacing the reference with consensus genomes impacts functional analyses, such as differential expressions of isoforms, genes, and splice junctions.

X-chromosome inactivation (XCI) is a random, permanent, and developmentally early epigenetic event that occurs during mammalian embryogenesis. We harness these features to investigate characteristics of early lineage specification events during human development. We initially assess the consistency of X-inactivation and establish a robust set of XCI-escape genes. By analyzing variance in XCI ratios across tissues and individuals, we find that XCI is shared across all tissues, suggesting that XCI is completed in the epiblast (in at least 6-16 cells) prior to specification of the germ layers. Additionally, we exploit tissue-specific variability to characterize the number of cells present during tissue-lineage commitment, ranging from approximately 20 cells in liver and whole blood tissues to 80 cells in brain tissues. By investigating the variability of XCI ratios using adult tissue, we characterize embryonic features of human XCI and lineage specification that are otherwise difficult to ascertain experimentally.

X-chromosome inactivation (XCI) is a random, permanent, and developmentally early epigenetic event that occurs during mammalian embryogenesis. We harness these features of XCI to investigate characteristics of early lineage specification events during human development. We initially assess the consistency of X-inactivation and establish a robust set of XCI-escape genes. By analyzing variance in XCI ratios across tissues and individuals, we find that XCI is completed prior to tissue specification and at a time when 6-16 cells are fated for all tissue lineages. Additionally, we exploit tissue specific variability to characterize the number of cells present at the time of each tissue’s lineage commitment, ranging from approximately 20 cells in liver and whole blood tissues to 80 cells in brain tissues. By investigating variance of XCI ratios using adult tissue, we resolve key features of human development otherwise difficult to ascertain experimentally and develop scalable methods easily applicable to future data.

Significance The identification of genes that are differentially expressed provides a molecular foothold onto biological questions of interest. Whether some genes are more likely to be differentially expressed than others, and to what degree, has never been assessed on a global scale. Here, we reanalyze more than 600 studies and find that knowledge of a gene’s prior probability of differential expression (DE) allows for accurate prediction of DE hit lists, regardless of the biological question. This result suggests redundancy in transcriptomics experiments that both informs gene set interpretation and highlights room for growth within the field., Differential expression (DE) is commonly used to explore molecular mechanisms of biological conditions. While many studies report significant results between their groups of interest, the degree to which results are specific to the question at hand is not generally assessed, potentially leading to inaccurate interpretation. This could be particularly problematic for metaanalysis where replicability across datasets is taken as strong evidence for the existence of a specific, biologically relevant signal, but which instead may arise from recurrence of generic processes. To address this, we developed an approach to predict DE based on an analysis of over 600 studies. A predictor based on empirical prior probability of DE performs very well at this task (mean area under the receiver operating characteristic curve, ∼0.8), indicating that a large fraction of DE hit lists are nonspecific. In contrast, predictors based on attributes such as gene function, mutation rates, or network features perform poorly. Genes associated with sex, the extracellular matrix, the immune system, and stress responses are prominent within the “DE prior.” In a series of control studies, we show that these patterns reflect shared biology rather than technical artifacts or ascertainment biases. Finally, we demonstrate the application of the DE prior to data interpretation in three use cases: (i) breast cancer subtyping, (ii) single-cell genomics of pancreatic islet cells, and (iii) metaanalysis of lung adenocarcinoma and renal transplant rejection transcriptomics. In all cases, we find hallmarks of generic DE, highlighting the need for nuanced interpretation of gene phenotypic associations.

The Human Reference Genome serves as the foundation for modern genomic analyses. However, in its present form, it does not adequately represent the vast genetic diversity of the human population. In this study, we explored the consensus genome as a potential successor of the current Reference genome and assessed its effect on the accuracy of RNA-seq read alignment. In order to find the best haploid genome representation, we constructed consensus genomes at the Pan-human, Super-population and Population levels, utilizing variant information from the 1000 Genomes project. Using personal haploid genomes as the ground truth, we compared mapping errors for real RNA-seq reads aligned to the consensus genomes versus the Reference genome. For reads overlapping homozygous variants, we found that the mapping error decreased by a factor of ~2-3 when the Reference was replaced with the Pan-human consensus genome. Interestingly, we also found that using more population-specific consensuses resulted in little to no increase over using the Pan-human consensus, suggesting a limit in the utility of incorporating more specific genomic variation. To assess the functional impact, we compared splice junction expression in the different genomes and found that the Pan-human consensus increases accuracy of splice junction quantification for hundreds of splice junctions.

X-linked chronic granulomatous disease (X-CGD) is a rare monogenetic immunodeficiency primarily affecting phagocytes. Precipitated by mutations in the CYBB gene, patients exhibit a compromised oxidative burst, leading to recurrent infections which can be life-threatening. Curiously, autoimmune manifestations are also common in patients and carriers. Here, exploiting the cell type-specific nature of this disorder, we characterize X-CGD on a transcriptional level using single-cell sequencing. Peripheral blood from 14 X-CGD probands and 10 carriers signed onto IRB approved protocol NCT00404560, as well as from 15 controls was sampled, and PBMCs and isolated monocytes were subjected to single-cell sequencing. Probands exhibited a strong differential expression signal relative to controls. This was composed of not only genes previously described to be up-regulated in X-CGD such as IFI27, and indeed an autoimmunity-associated broader type I interferon response, but also previously undescribed genes involved in monocyte function (ARG1), antimicrobial proteins (CAMP, SLPI), and inflammasome components (AIM2). Surprisingly, expression variability was not greater in carriers relative to probands or controls, indicating a lack of cell autonomous effects from the deletion of CYBB. Interestingly, aggregate expression of differentially expressed genes in the probands was able to classify carriers from sex-matched controls with high accuracy (AUROC = 0.92), indicating the presence of an X-CGD-specific gene signature. This gene signature was also strongly co-expressed across 17 chordate species, pointing towards the disruption of ancestral pathways important in antimicrobial immunity in X-CGD probands and carriers. This work was partially supported by a Swiss National Science Foundation fellowship to S.M.

Co-expression analysis has provided insight into gene function in organisms from Arabidopsis to Zebrafish. Comparison across species has the potential to enrich these results, for example by prioritizing among candidate human disease genes based on their network properties, or by finding alternative model systems where their co-expression is conserved. Here, we present CoCoCoNet as a tool for identifyingconserved gene modules andcomparingco-expressionnetworks. CoCoCoNet is a resource for both data and methods, providing gold-standard networks and sophisticated tools for on-the-fly comparative analyses across 14 species. We show how CoCoCoNet can be used in two use cases. In the first, we demonstrate deep conservation of a nucleolus gene module across very divergent organisms, and in the second, we show how the heterogeneity of autism mechanisms in humans can be broken down by functional groups, and translated to model organisms. CoCoCoNet is free to use and available to all athttps://milton.cshl.edu/CoCoCoNet, with data and R scripts available atftp://milton.cshl.edu/data.

Genetic variation, epigenetic regulation and major environmental stimuli are key contributors to phenotypic variation, but the influence of minor perturbations or “noise” has been difficult to assess in mammals. In this work, we uncover one major axis of random variation with a large and permanent influence: developmental stochasticity. By assaying the transcriptome of wild monozygotic quadruplets of the nine-banded armadillo, we find that persistent changes occur early in development, and these give rise to clear transcriptional signatures which uniquely characterize individuals relative to siblings. Comparing these results to human twins, we find the transcriptional signatures which define individuals exhibit conserved co-expression, suggesting a substantial fraction of phenotypic and disease discordance within mammals arises from developmental stochasticity.One sentence summaryLongitudinal gene expression in identical armadillo quadruplets reveals a major role for developmental stochasticity.

Cardiac electrical activity is controlled by the carefully orchestrated activity of more than a dozen different ion conductances. Yet, there is considerable variability in cardiac ion channel expression levels both within and between subjects. In this study we tested the hypothesis that variations in ion channel expression between individuals are not random but rather there are modules of co-expressed genes and that these modules make electrical signaling in the heart more robust.Meta-analysis of 3653 public RNA-Seq datasets identified a strong correlation between expression of CACNA1C (L-type calcium current, ICaL) and KCNH2 (rapid delayed rectifier K+ current, IKr), which was verified in mRNA extracted from human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CM). In silico modeling, validated with functional measurements in hiPSC-CM, indicates that the co-expression of CACNA1C and KCNH2 limits the variability in action potential duration and reduces susceptibility to early afterdepolarizations, a surrogate marker for pro-arrhythmia.Impact StatementCoexpressed levels of potassium and calcium ion channel genes in the heart encode more robust cardiac electrophysiology and provide insights into genetic basis of arrhythmic risk

Motivation: Gene networks have become a central tool in the analysis of genomic data but are widely regarded as hard to interpret. This has motivated a great deal of comparative evaluation and research into best practices. We explore the possibility that this may lead to overfitting in the field as a whole. Results: We construct a model of ‘research communities’ sampling from real gene network data and machine learning methods to characterize performance trends. Our analysis reveals an important principle limiting the value of replication, namely that targeting it directly causes ‘easy’ or uninformative replication to dominate analyses. We find that when sampling across network data and algorithms with similar variability, the relationship between replicability and accuracy is positive (Spearman’s correlation, rs ∼0.33) but where no such constraint is imposed, the relationship becomes negative for a given gene function (rs ∼ −0.13). We predict factors driving replicability in some prior analyses of gene networks and show that they are unconnected with the correctness of the original result, instead reflecting replicable biases. Without these biases, the original results also vanish replicably. We show these effects can occur quite far upstream in network data and that there is a strong tendency within protein–protein interaction data for highly replicable interactions to be associated with poor quality control. Availability and implementation: Algorithms, network data and a guide to the code available at: https://github.com/wimverleyen/AggregateGeneFunctionPrediction. Contact: jgillis@cshl.edu Supplementary information: Supplementary data are available at Bioinformatics online.

Chromatin accessibility provides an important window into the regulation of gene expression. Recently, the Assay of Transposase Accessible Chromatin with sequencing (ATAC-seq) was developed to profile genome-wide chromatin accessibility. In this work we analyzed 193 ATAC-seq samples from 12 studies to determine the robustness of peaks, as well as their specificity across studies. We employ a read-downsampling approach and find that even samples which have high average read-depth yield poorly powered peaks using conventional pipelines. Finding that many peaks are sensitive to slight perturbations in read-sampling, we develop an approach to improve the robustness of peak signals and apply this novel method to detect differential accessibility between the amygdala and cortex. Most peak calling results heavily and unreasonably depend on the absolute number of reads, easily yielding artifacts when differential peak evaluations are performed. These problems can be ameliorated by resampling reads to determine the robust aggregate result.

Single-cell RNA-sequencing (scRNA-seq) technology provides a new avenue to discover and characterize cell types; however, the experiment-specific technical biases and analytic variability inherent to current pipelines may undermine its replicability. Meta-analysis is further hampered by the use of ad hoc naming conventions. Here we demonstrate our replication framework, MetaNeighbor, that quantifies the degree to which cell types replicate across datasets, and enables rapid identification of clusters with high similarity. We first measure the replicability of neuronal identity, comparing results across eight technically and biologically diverse datasets to define best practices for more complex assessments. We then apply this to novel interneuron subtypes, finding that 24/45 subtypes have evidence of replication, which enables the identification of robust candidate marker genes. Across tasks we find that large sets of variably expressed genes can identify replicable cell types with high accuracy, suggesting a general route forward for large-scale evaluation of scRNA-seq data., Single cell RNA-sequencing analysis poses challenges in replication due to technical biases and analytic variability among bioinformatics pipelines. Here, Crow et al develop MetaNeighbor for measuring cell-type replication across datasets, and use it to identify marker genes for neuron subtypes with evidence of replication.

Many tools are available for RNA-seq alignment and expression quantification, with comparative value being hard to establish. Benchmarking assessments often highlight methods’ good performance, but are focused on either model data or fail to explain variation in performance. This leaves us to ask, what is the most meaningful way to assess different alignment choices? And importantly, where is there room for progress? In this work, we explore the answers to these two questions by performing an exhaustive assessment of the STAR aligner. We assess STAR’s performance across a range of alignment parameters using common metrics, and then on biologically focused tasks. We find technical metrics such as fraction mapping or expression profile correlation to be uninformative, capturing properties unlikely to have any role in biological discovery. Surprisingly, we find that changes in alignment parameters within a wide range have little impact on both technical and biological performance. Yet, when performance finally does break, it happens in difficult regions, such as X-Y paralogs and MHC genes. We believe improved reporting by developers will help establish where results are likely to be robust or fragile, providing a better baseline to establish where methodological progress can still occur.

Single cell RNA-sequencing technology (scRNA-seq) provides a new avenue to discover and characterize cell types, but the experiment-specific technical biases and analytic variability inherent to current pipelines may undermine the replicability of these studies. Meta-analysis of rapidly accumulating data is further hampered by the use of ad hoc naming conventions. Here we demonstrate our replication framework, MetaNeighbor, that allows researchers to quantify the degree to which cell types replicate across datasets, and to rapidly identify clusters with high similarity for further testing. We first measure the replicability of neuronal identity by comparing more than 13 thousand individual scRNA-seq transcriptomes, sampling with high specificity from within the data to define a range of robust practices. We then assess cross-dataset evidence for novel cortical interneuron subtypes identified by scRNA-seq and find that 24/45 cortical interneuron subtypes have evidence of replication in at least one other study. Identifying these putative replicates allows us to re-analyze the data for differential expression and provide lists of robust candidate marker genes. Across tasks we find that large sets of variably expressed genes can identify replicable cell types and subtypes with high accuracy, suggesting a general route forward for large-scale evaluation of scRNA-seq data.

SummaryIn characterizing a disease, it is common to search for dysfunctional genes by assaying the transcriptome. The resulting differentially expressed genes are typically assessed for shared features, such as functional annotation or co-expression. While useful, the reliability of these systems methods is hard to evaluate. To better understand shared disease signals, we assess their replicability by first looking at gene-level recurrence and then pathway-level recurrence along with co-expression signals across six pedigrees of a rare homogeneous X-linked disorder, TAF1 syndrome. We find most differentially expressed genes are not recurrent between pedigrees, making functional enrichment largely distinct in each pedigree. However, we find two highly recurrent “functional outliers” (CACNA1I and IGFBP3), genes acting atypically with respect to co-expression and therefore absent from a systems-level assessment. We show this occurs in re-analysis of Huntington’s disease, Parkinson’s disease and schizophrenia. Our results suggest a significant role for genes easily missed in systems approaches.

Networks constructed from aggregated protein-protein interaction data are commonplace in biology. But the studies these data are derived from were conducted with their own hypotheses and foci. Focusing on data from budding yeast present in BioGRID, we determine that many of the downstream signals present in network data are significantly impacted by biases in the original data. We determine the degree to which selection bias in favor of biologically interesting bait proteins goes down with study size, while we also find that promiscuity in prey contributes more substantially in larger studies. We analyze interaction studies over time with respect to data in the Gene Ontology and find that reproducibly observed interactions are less likely to favor multifunctional proteins. We find that strong alignment between co-expression and protein-protein interaction data occurs only for extreme co-expression values, and use this data to suggest candidates for targets likely to reveal novel biology in follow-up studies.Protein-protein interaction data finds particularly heavy use in the interpretation of disease-causal variants. In principle, network data allows researchers to find novel commonalities among candidate genes. In this study, we detail several of the most salient biases contributing to aggregated protein-protein interaction databases. We find strong evidence for the role of selection and laboratory biases. Many of these effects contribute to the commonalities researchers find for disease genes. In order for characterization of disease genes and their interactions to not simply be an artifact of researcher preference, it is imperative to identify data biases explicitly. Based on this, we also suggest ways to move forward in producing candidates less influenced by prior knowledge. This article is part of a Special Issue entitled: Can Proteomics Fill the Gap Between Genomics and Phenotypes?

Gene set analysis, which translates gene lists into enriched functions, is among the most common bioinformatic methods. Yet few would advocate taking the results at face value. Not only is there no agreement on the algorithms themselves, there is no agreement on how to benchmark them. In this paper, we evaluate the robustness and uniqueness of enrichment results as a means of assessing methods even where correctness is unknown. We show that heavily annotated (“multifunctional”) genes are likely to appear in genomics study results and drive the generation of biologically non-specific enrichment results as well as highly fragile significances. By providing a means of determining where enrichment analyses report non-specific and non-robust findings, we are able to assess where we can be confident in their use. We find significant progress in recent bias correction methods for enrichment and provide our own software implementation. Our approach can be readily adapted to any pre-existing package.

SummaryEvaluating gene networks with respect to known biology is a common task but often a computationally costly one. Many computational experiments are difficult to apply exhaustively in network analysis due to run-times. To permit high-throughput analysis of gene networks, we have implemented a set of very efficient tools to calculate functional properties in networks based on guilt-by-association methods. EGAD (Extending ‘Guilt-by-Association’ by Degree) allows gene networks to be evaluated with respect to hundreds or thousands of gene sets. The methods predict novel members of gene groups, assess how well a gene network groups known sets of genes, and determines the degree to which generic predictions drive performance. By allowing fast evaluations, whether of random sets or real functional ones, EGAD provides the user with an assessment of performance which can easily be used in controlled evaluations across many parameters.Availability and ImplementationThe software package is freely available at https://github.com/sarbal/EGAD and implemented for use in R and Matlab. The package is also freely available under the LGPL license from the Bioconductor web site (http://bioconductor.org).ContactJGillis@cshl.eduSupplementary informationSupplementary data are available at Bioinformatics online and the full manual at http://gillislab.labsites.cshl.edu/software/egad-extending-guilt-by-association-by-degree/.

BackgroundDisagreements over genetic signatures associated with disease have been particularly prominent in the field of psychiatric genetics, creating a sharp divide between disease burdens attributed to common and rare variation, with study designs independently targeting each. Meta-analysis within each of these study designs is routine, whether using raw data or summary statistics, but combining results across study designs is atypical. However, tests of functional convergence are used across all study designs, where candidate gene sets are assessed for overlaps with previously known properties. This suggests one possible avenue for combining not study data, but the functional conclusions that they reach.MethodIn this work, we test for functional convergence in autism spectrum disorder (ASD) across different study types, and specifically whether the degree to which a gene is implicated in autism is correlated with the degree to which it drives functional convergence. Because different study designs are distinguishable by their differences in effect size, this also provides a unified means of incorporating the impact of study design into the analysis of convergence.ResultsWe detected remarkably significant positive trends in aggregate (p < 2.2e-16) with 14 individually significant properties (FDR(FMRP)interactor enrichment (FDR 0.003). We are also able to detect novel technical effects and we see that network enrichment from protein-protein interaction data is heavily confounded with study design, arising readily in control data.ConclusionsWe see a convergent functional signal for a subset of known and novel functions in ASD from all sources of genetic variation. Meta-analytic approaches explicitly accounting for different study designs can be adapted to other diseases to discover novel functional associations and increase statistical power.

Background Co-expression networks have been a useful tool for functional genomics, providing important clues about the cellular and biochemical mechanisms that are active in normal and disease processes. However, co-expression analysis is often treated as a black box with results being hard to trace to their basis in the data. Here, we use both published and novel single-cell RNA sequencing (RNA-seq) data to understand fundamental drivers of gene-gene connectivity and replicability in co-expression networks. Results We perform the first major analysis of single-cell co-expression, sampling from 31 individual studies. Using neighbor voting in cross-validation, we find that single-cell network connectivity is less likely to overlap with known functions than co-expression derived from bulk data, with functional variation within cell types strongly resembling that also occurring across cell types. To identify features and analysis practices that contribute to this connectivity, we perform our own single-cell RNA-seq experiment of 126 cortical interneurons in an experimental design targeted to co-expression. By assessing network replicability, semantic similarity and overall functional connectivity, we identify technical factors influencing co-expression and suggest how they can be controlled for. Many of the technical effects we identify are expression-level dependent, making expression level itself highly predictive of network topology. We show this occurs generally through re-analysis of the BrainSpan RNA-seq data. Conclusions Technical properties of single-cell RNA-seq data create confounds in co-expression networks which can be identified and explicitly controlled for in any supervised analysis. This is useful both in improving co-expression performance and in characterizing single-cell data in generally applicable terms, permitting cross-laboratory comparison within a common framework. Electronic supplementary material The online version of this article (doi:10.1186/s13059-016-0964-6) contains supplementary material, which is available to authorized users.

The expansion of protein-ligand annotation databases has enabled large-scale networking of proteins by ligand similarity. These ligand-based protein networks, which implicitly predict the ability of neighboring proteins to bind related ligands, may complement biologically-oriented gene networks, which are used to predict functional or disease relevance. To quantify the degree to which such ligand-based protein associations might complement functional genomic associations, including sequence similarity, physical protein-protein interactions, co-expression, and disease gene annotations, we calculated a network based on the Similarity Ensemble Approach (SEA: sea.docking.org), where protein neighbors reflect the similarity of their ligands. We also measured the similarity with functional genomic networks over a common set of 1,131 genes, and found that the networks had only small overlaps, which were significant only due to the large scale of the data. Consistent with the view that the networks contain different information, combining them substantially improved Molecular Function prediction within GO (from AUROC~0.63-0.75 for the individual data modalities to AUROC~0.8 in the aggregate). We investigated the boost in guilt-by-association gene function prediction when the networks are combined and describe underlying properties that can be further exploited.

Background Coronary artery disease (CAD), one of the leading causes of death globally, is influenced by both environmental and genetic risk factors. Gene-centric genome-wide association studies (GWAS) involving cases and controls have been remarkably successful in identifying genetic loci contributing to CAD. Modern in silico platforms, such as candidate gene prediction tools, permit a systematic analysis of GWAS data to identify candidate genes for complex diseases like CAD. Subsequent integration of drug-target data from drug databases with the predicted candidate genes can potentially identify novel therapeutics suitable for repositioning towards treatment of CAD. Methods Previously, we were able to predict 264 candidate genes and 104 potential therapeutic targets for CAD using Gentrepid (http://www.gentrepid.org), a candidate gene prediction platform with two bioinformatic modules to reanalyze Wellcome Trust Case-Control Consortium GWAS data. In an expanded study, using five bioinformatic modules on the same data, Gentrepid predicted 647 candidate genes and successfully replicated 55% of the candidate genes identified by the more powerful CARDIoGRAMplusC4D consortium meta-analysis. Hence, Gentrepid was capable of enhancing lower quality genotype-phenotype data, using an independent knowledgebase of existing biological data. Here, we used our methodology to integrate drug data from three drug databases: the Therapeutic Target Database, PharmGKB and Drug Bank, with the 647 candidate gene predictions from Gentrepid. We utilized known CAD targets, the scientific literature, existing drug data and the CARDIoGRAMplusC4D meta-analysis study as benchmarks to validate Gentrepid predictions for CAD. Results Our analysis identified a total of 184 predicted candidate genes as novel therapeutic targets for CAD, and 981 novel therapeutics feasible for repositioning in clinical trials towards treatment of CAD. The benchmarks based on known CAD targets and the scientific literature showed that our results were significant (p < 0.05). Conclusions We have demonstrated that available drugs may potentially be repositioned as novel therapeutics for the treatment of CAD. Drug repositioning can save valuable time and money spent on preclinical and phase I clinical studies.

In addition to detecting novel transcripts and higher dynamic range, a principal claim for RNA-sequencing has been greater replicability, typically measured in sample-sample correlations of gene expression levels. Through a re-analysis of ENCODE data, we show that replicability of transcript abundances will provide misleading estimates of the replicability of conditional variation in transcript abundances (i.e., most expression experiments). Heuristics which implicitly address this problem have emerged in quality control measures to obtain ‘good’ differential expression results. However, these methods involve strict filters such as discarding low expressing genes or using technical replicates to remove discordant transcripts, and are costly or simply ad hoc. As an alternative, we model gene-level replicability of differential activity using co-expressing genes. We find that sets of housekeeping interactions provide a sensitive means of estimating the replicability of expression changes, where the co-expressing pair can be regarded as pseudo-replicates of one another. We model the effects of noise that perturbs a gene’s expression within its usual distribution of values and show that perturbing expression by only 5% within that range is readily detectable (AUROC~0.73). We have made our method available as a set of easily implemented R scripts., Author Summary RNA-sequencing has become a popular means to detect the expression levels of genes. However, quality control is still challenging, requiring both extreme measures and rules which are set in stone from extensive previous analysis. Instead of relying on these rules, we show that co-expression can be used to measure biological replicability with extremely high precision. Co-expression is a well-studied phenomenon in which two genes that are known to form a functional unit are also expressed at similar levels, and change in similar ways across conditions. Using this concept, we can detect how well an experiment replicates by measuring how well it has retained the co-expression pattern across defined gene-pairs. We do this by measuring how easy it is to detect a sample to which some noise has been added. We show this method is a useful tool for quality control.

Motivation: Network-based gene function inference methods have proliferated in recent years, but measurable progress remains elusive. We wished to better explore performance trends by controlling data and algorithm implementation, with a particular focus on the performance of aggregate predictions. Results: Hypothesizing that popular methods would perform well without hand-tuning, we used well-characterized algorithms to produce verifiably ‘untweaked’ results. We find that most state-of-the-art machine learning methods obtain ‘gold standard’ performance as measured in critical assessments in defined tasks. Across a broad range of tests, we see close alignment in algorithm performances after controlling for the underlying data being used. We find that algorithm aggregation provides only modest benefits, with a 17% increase in area under the ROC (AUROC) above the mean AUROC. In contrast, data aggregation gains are enormous with an 88% improvement in mean AUROC. Altogether, we find substantial evidence to support the view that additional algorithm development has little to offer for gene function prediction. Availability and implementation: The supplementary information contains a description of the algorithms, the network data parsed from different biological data resources and a guide to the source code (available at: http://gillislab.cshl.edu/supplements/). Contact: jgillis@cshl.edu

Motivation: RNA-seq co-expression analysis is in its infancy and reasonable practices remain poorly defined. We assessed a variety of RNA-seq expression data to determine factors affecting functional connectivity and topology in co-expression networks. Results: We examine RNA-seq co-expression data generated from 1970 RNA-seq samples using a Guilt-By-Association framework, in which genes are assessed for the tendency of co-expression to reflect shared function. Minimal experimental criteria to obtain performance on par with microarrays were >20 samples with read depth >10 M per sample. While the aggregate network constructed shows good performance (area under the receiver operator characteristic curve ∼0.71), the dependency on number of experiments used is nearly identical to that present in microarrays, suggesting thousands of samples are required to obtain ‘gold-standard’ co-expression. We find a major topological difference between RNA-seq and microarray co-expression in the form of low overlaps between hub-like genes from each network due to changes in the correlation of expression noise within each technology. Contact: jgillis@cshl.edu or sballouz@cshl.edu Supplementary information: Networks are available at: http://gillislab.labsites.cshl.edu/supplements/rna-seq-networks/ and supplementary data are available at Bioinformatics online.

Background Human genome sequencing has enabled the association of phenotypes with genetic loci, but our ability to effectively translate this data to the clinic has not kept pace. Over the past 60 years, pharmaceutical companies have successfully demonstrated the safety and efficacy of over 1,200 novel therapeutic drugs via costly clinical studies. While this process must continue, better use can be made of the existing valuable data. In silico tools such as candidate gene prediction systems allow rapid identification of disease genes by identifying the most probable candidate genes linked to genetic markers of the disease or phenotype under investigation. Integration of drug-target data with candidate gene prediction systems can identify novel phenotypes which may benefit from current therapeutics. Such a drug repositioning tool can save valuable time and money spent on preclinical studies and phase I clinical trials. Methods We previously used Gentrepid (http://www.gentrepid.org) as a platform to predict 1,497 candidate genes for the seven complex diseases considered in the Wellcome Trust Case-Control Consortium genome-wide association study; namely Type 2 Diabetes, Bipolar Disorder, Crohn's Disease, Hypertension, Type 1 Diabetes, Coronary Artery Disease and Rheumatoid Arthritis. Here, we adopted a simple approach to integrate drug data from three publicly available drug databases: the Therapeutic Target Database, the Pharmacogenomics Knowledgebase and DrugBank; with candidate gene predictions from Gentrepid at the systems level. Results Using the publicly available drug databases as sources of drug-target association data, we identified a total of 428 candidate genes as novel therapeutic targets for the seven phenotypes of interest, and 2,130 drugs feasible for repositioning against the predicted novel targets. Conclusions By integrating genetic, bioinformatic and drug data, we have demonstrated that currently available drugs may be repositioned as novel therapeutics for the seven diseases studied here, quickly taking advantage of prior work in pharmaceutics to translate ground-breaking results in genetics to clinical treatments.

Current single‐locus‐based analyses and candidate disease gene prediction methodologies used in genome‐wide association studies (GWAS) do not capitalize on the wealth of the underlying genetic data, nor functional data available from molecular biology. Here, we analyzed GWAS data from the Wellcome Trust Case Control Consortium (WTCCC) on coronary artery disease (CAD). Gentrepid uses a multiple‐locus‐based approach, drawing on protein pathway‐ or domain‐based data to make predictions. Known disease genes may be used as additional information (seeded method) or predictions can be based entirely on GWAS single nucleotide polymorphisms (SNPs) (ab initio method). We looked in detail at specific predictions made by Gentrepid for CAD and compared these with known genetic data and the scientific literature. Gentrepid was able to extract known disease genes from the candidate search space and predict plausible novel disease genes from both known and novel WTCCC‐implicated loci. The disease gene candidates are consistent with known biological information. The results demonstrate that this computational approach is feasible and a valuable discovery tool for geneticists.

Background Candidate disease gene prediction is a rapidly developing area of bioinformatics research with the potential to deliver great benefits to human health. As experimental studies detecting associations between genetic intervals and disease proliferate, better bioinformatic techniques that can expand and exploit the data are required. Description Gentrepid is a web resource which predicts and prioritizes candidate disease genes for both Mendelian and complex diseases. The system can take input from linkage analysis of single genetic intervals or multiple marker loci from genome-wide association studies. The underlying database of the Gentrepid tool sources data from numerous gene and protein resources, taking advantage of the wealth of biological information available. Using known disease gene information from OMIM, the system predicts and prioritizes disease gene candidates that participate in the same protein pathways or share similar protein domains. Alternatively, using an ab initio approach, the system can detect enrichment of these protein annotations without prior knowledge of the phenotype. Conclusions The system aims to integrate the wealth of protein information currently available with known and novel phenotype/genotype information to acquire knowledge of biological mechanisms underpinning disease. We have updated the system to facilitate analysis of GWAS data and the study of complex diseases. Application of the system to GWAS data on hypertension using the ICBP data is provided as an example. An interesting prediction is a ZIP transporter additional to the one found by the ICBP analysis. The webserver URL is https://www.gentrepid.org/.

Despite increasing sequencing capacity, genetic disease investigation still frequently results in the identification of loci containing multiple candidate disease genes that need to be tested for involvement in the disease. This process can be expedited by prioritizing the candidates prior to testing. Over the last decade, a large number of computational methods and tools have been developed to assist the clinical geneticist in prioritizing candidate disease genes. In this chapter, we give an overview of computational tools that can be used for this purpose, all of which are freely available over the web.

Genes encoding proteins in a common pathway are often found near each other along bacterial chromosomes. Several explanations have been proposed to account for the evolution of these structures. For instance, natural selection may directly favour gene clusters through a variety of mechanisms, such as increased efficiency of coregulation. An alternative and controversial hypothesis is the selfish operon model, which asserts that clustered arrangements of genes are more easily transferred to other species, thus improving the prospects for survival of the cluster. According to another hypothesis (the persistence model), genes that are in close proximity are less likely to be disrupted by deletions. Here we develop computational models to study the conditions under which gene clusters can evolve and persist. First, we examine the selfish operon model by re-implementing the simulation and running it under a wide range of conditions. Second, we introduce and study a Moran process in which there is natural selection for gene clustering and rearrangement occurs by genome inversion events. Finally, we develop and study a model that includes selection and inversion, which tracks the occurrence and fixation of rearrangements. Surprisingly, gene clusters fail to evolve under a wide range of conditions. Factors that promote the evolution of gene clusters include a low number of genes in the pathway, a high population size, and in the case of the selfish operon model, a high horizontal transfer rate. The computational analysis here has shown that the evolution of gene clusters can occur under both direct and indirect selection as long as certain conditions hold. Under these conditions the selfish operon model is still viable as an explanation for the evolution of gene clusters., Author Summary Genes involved in a common pathway or function are frequently found near each other on bacterial chromosomes. A number of hypotheses have been previously presented to explain this observation. A particularly influential theory is the selfish operon model, which posits that horizontal transfer could promote gene clustering by favouring transfer of arrangements of genes that are close together. Subsequent theoretical development and analysis of genomic data have contributed to the debate about the plausibility of this model. Here, by re-examining the evolutionary dynamics of gene clusters, we provide and discuss conditions under which gene clusters can evolve. We find that first, some form of bias for clustering is required for clusters to evolve. This bias can be in the form of bias in horizontal transfer towards genes that are close together, or direct natural selection for gene proximity. Our computational work does not present a theoretical obstacle to the selfish operon model as a possible explanation for the evolution of gene clusters.

Background Automated candidate gene prediction systems allow geneticists to hone in on disease genes more rapidly by identifying the most probable candidate genes linked to the disease phenotypes under investigation. Here we assessed the ability of eight different candidate gene prediction systems to predict disease genes in intervals previously associated with type 2 diabetes by benchmarking their performance against genes implicated by recent genome-wide association studies. Results Using a search space of 9556 genes, all but one of the systems pruned the genome in favour of genes associated with moderate to highly significant SNPs. Of the 11 genes associated with highly significant SNPs identified by the genome-wide association studies, eight were flagged as likely candidates by at least one of the prediction systems. A list of candidates produced by a previous consensus approach did not match any of the genes implicated by 706 moderate to highly significant SNPs flagged by the genome-wide association studies. We prioritized genes associated with medium significance SNPs. Conclusion The study appraises the relative success of several candidate gene prediction systems against independent genetic data. Even when confronted with challengingly large intervals, the candidate gene prediction systems can successfully select likely disease genes. Furthermore, they can be used to filter statistically less-well-supported genetic data to select more likely candidates. We suggest consensus approaches fail because they penalize novel predictions made from independent underlying databases. To realize their full potential further work needs to be done on prioritization and annotation of genes.