MOTIVATION: Droplet-based single-cell RNA sequencing (scRNA-seq) is widely used in biomedical research for interrogating the transcriptomes of single cells on a large scale. Pooling and processing cells from different samples together can reduce costs and batch effects. To pool cells, they are often first labeled with hashtag oligonucleotides (HTOs). These HTOs are sequenced alongside the cells' RNA in the droplets and subsequently used to computationally assign each droplet to its sample of origin, a process referred to as demultiplexing. Accurate demultiplexing is crucial but can be challenging due to background HTOs, low-quality cells/cell debris, and multiplets. RESULTS: A new demultiplexing method based on negative binomial regression mixture models is introduced. The method, called demuxmix, implements two significant improvements. First, demuxmix's probabilistic classification framework provides error probabilities for droplet assignments that can be used to discard uncertain droplets and inform about the quality of the HTO data and the success of the demultiplexing process. Second, demuxmix utilizes the positive association between detected genes in the RNA library and HTO counts to explain parts of the variance in the HTO data resulting in improved droplet assignments. The improved performance of demuxmix compared with existing demultiplexing methods is assessed using real and simulated data. Finally, the feasibility of accurately demultiplexing experimental designs where non-labeled cells are pooled with labeled cells is demonstrated. AVAILABILITY AND IMPLEMENTATION: R/Bioconductor package demuxmix (https://doi.org/doi:10.18129/B9.bioc.demuxmix).

Biomedical research studies have generated large multi-omic datasets to study complex diseases like Alzheimer's disease (AD). An important aim of these studies is the identification of candidate genes that demonstrate congruent disease-related alterations across the different data types measured by the study. We developed a new method to detect such candidate genes in large multi-omic case-control studies that measure multiple data types in the same set of samples. The method is based on a gene-centric integrative coefficient quantifying to what degree consistent differences are observed in the different data types. For statistical inference, a Bayesian hierarchical model is used to study the distribution of the integrative coefficient. The model employs a conditional autoregressive prior to integrate a functional gene network and to share information between genes known to be functionally related. We applied the method to an AD dataset consisting of histone acetylation, DNA methylation, and RNA transcription data from human cortical tissue samples of 233 subjects, and we detected 816 genes with consistent differences between persons with AD and controls. The findings were validated in protein data and in RNA transcription data from two independent AD studies. Finally, we found three subnetworks of jointly dysregulated genes within the functional gene network which capture three distinct biological processes: myeloid cell differentiation, protein phosphorylation and synaptic signaling. Further investigation of the myeloid network indicated an upregulation of this network in early stages of AD prior to accumulation of hyperphosphorylated tau and suggested that increased CSF1 transcription in astrocytes may contribute to microglial activation in AD. Thus, we developed a method that integrates multiple data types and external knowledge of gene function to detect candidate genes, applied the method to an AD dataset, and identified several disease-related genes and processes demonstrating the usefulness of the integrative approach.

MOTIVATION: Genes showing congruent differences in several genomic variables between two biological conditions are crucial to unravel causalities behind phenotypes of interest. Detecting such genes is important in biomedical research, e.g. when identifying genes responsible for cancer development. Small sample sizes common in next-generation sequencing studies are a key challenge, and there are still only very few statistical methods to analyze more than two genomic variables in an integrative, model-based way. Here, we present a novel bioinformatics approach to detect congruent differences between two biological conditions in a larger number of different measurements such as various epigenetic marks or mRNA transcript levels. RESULTS: We propose a coefficient quantifying the degree to which genes present consistent alterations in multiple (more than two) genomic variables when comparing samples presenting a condition of interest (e.g. cancer) to a reference group. A hierarchical Bayesian model is employed to assess uncertainty on a gene level, incorporating information on functional relationships between genes. We demonstrate the approach on different data sets containing RNA-seq gene transcripton and up to four ChIP-seq histone modification measurements. Both the coefficient-based ranking and the inference based on the model lead to a plausible prioritizing of candidate genes when analyzing multiple genomic variables. AVAILABILITY AND IMPLEMENTATION: BUGS code in the Supplement. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

In the biology of tissue development and diseases, DNA methylation plays an important role. For a deeper understanding, it is crucial to accurately compare DNA methylation patterns between groups of samples representing different conditions. A widely used method to investigate DNA methylation in the CpG context is bisulfite sequencing, which produces data on the single-nucleotide scale. While there are benefits to analyzing CpG sites on a basepair level, there are both biological and statistical reasons to test entire genomic regions for differential methylation. However, the analysis of DNA methylation is hampered by the lack of best practice standards. Here, we compared multiple approaches for testing predefined genomic regions for differential DNA methylation in bisulfite sequencing data. Nine methods were evaluated: BiSeq, COHCAP, Goeman's Global Test, Limma, methylKit/eDMR, RADMeth and three log-linear regression approaches with different distribution assumptions. We applied these methods to simulated data and determined their sensitivity and specificity. This revealed performance differences, which were also seen when applied to real data. Methods that first test single CpG sites and then test regions based on transformed CpG-wise P-values performed better than methods that summarize methylation levels or raw reads. Interestingly, smoothing of methylation levels had a negligible impact. In particular, Global Test, BiSeq and RADMeth/z-test outperformed the other methods we evaluated, providing valuable guidance for more accurate analysis of DNA methylation.

MOTIVATION: Histone modifications are a key epigenetic mechanism to activate or repress the transcription of genes. Datasets of matched transcription data and histone modification data obtained by ChIP-seq exist, but methods for integrative analysis of both data types are still rare. Here, we present a novel bioinformatics approach to detect genes that show different transcript abundances between two conditions putatively caused by alterations in histone modification. RESULTS: We introduce a correlation measure for integrative analysis of ChIP-seq and gene transcription data measured by RNA sequencing or microarrays and demonstrate that a proper normalization of ChIP-seq data is crucial. We suggest applying Bayesian mixture models of different types of distributions to further study the distribution of the correlation measure. The implicit classification of the mixture models is used to detect genes with differences between two conditions in both gene transcription and histone modification. The method is applied to different datasets, and its superiority to a naive separate analysis of both data types is demonstrated. AVAILABILITY AND IMPLEMENTATION: R/Bioconductor package epigenomix. CONTACT: [email protected] Supplementary information: Supplementary data are available at Bioinformatics online.

A variety of genome-wide profiling techniques are available to investigate complementary aspects of genome structure and function. Integrative analysis of heterogeneous data sources can reveal higher level interactions that cannot be detected based on individual observations. A standard integration task in cancer studies is to identify altered genomic regions that induce changes in the expression of the associated genes based on joint analysis of genome-wide gene expression and copy number profiling measurements. In this review, we highlight common approaches to genomic data integration and provide a transparent benchmarking procedure to quantitatively compare method performances in cancer gene prioritization. Algorithms, data sets and benchmarking results are available at http://intcomp.r-forge.r-project.org.

MOTIVATION: Bisulfite sequencing is currently the gold standard to obtain genome-wide DNA methylation profiles in eukaryotes. In contrast to the rapid development of appropriate pre-processing and alignment software, methods for analyzing the resulting methylation profiles are relatively limited so far. For instance, an appropriate pipeline to detect DNA methylation differences between cancer and control samples is still required. RESULTS: We propose an algorithm that detects significantly differentially methylated regions in data obtained by targeted bisulfite sequencing approaches, such as reduced representation bisulfite sequencing. In a first step, this approach tests all target regions for methylation differences by taking spatial dependence into account. A false discovery rate procedure controls the expected proportion of incorrectly rejected regions. In a second step, the significant target regions are trimmed to the actually differentially methylated regions. This hierarchical procedure detects differentially methylated regions with increased power compared with existing methods. AVAILABILITY: R/Bioconductor package BiSeq. SUPPLEMENTARY INFORMATION: Supplementary Data are available at Bioinformatics online.

UNLABELLED: The R453Plus1Toolbox is an R/Bioconductor package for the analysis of 454 Sequencing data. Projects generated with Roche's data analysis software can be imported into R allowing advanced and customized analyses within the R/Bioconductor environment for sequencing data. Several methods were implemented extending the current functionality of Roche's software. These extensions include methods for quality assurance and annotation of detected variants. Further, a pipeline for the detection of structural variants, e.g. balanced chromosomal translocations, is provided. AVAILABILITY: The R453Plus1Toolbox is implemented in R and available at http://www.bioconductor.org/. A vignette outlining typical workflows is included in the package. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Visualization of DNA microarray data in two or three dimensional spaces is an important exploratory analysis step in order to detect quality issues or to generate new hypotheses. Principal Component Analysis (PCA) is a widely used linear method to define the mapping between the high-dimensional data and its low-dimensional representation. During the last decade, many new nonlinear methods for dimension reduction have been proposed, but it is still unclear how well these methods capture the underlying structure of microarray gene expression data. In this study, we assessed the performance of the PCA approach and of six nonlinear dimension reduction methods, namely Kernel PCA, Locally Linear Embedding, Isomap, Diffusion Maps, Laplacian Eigenmaps and Maximum Variance Unfolding, in terms of visualization of microarray data. RESULTS: A systematic benchmark, consisting of Support Vector Machine classification, cluster validation and noise evaluations was applied to ten microarray and several simulated datasets. Significant differences between PCA and most of the nonlinear methods were observed in two and three dimensional target spaces. With an increasing number of dimensions and an increasing number of differentially expressed genes, all methods showed similar performance. PCA and Diffusion Maps responded less sensitive to noise than the other nonlinear methods. CONCLUSIONS: Locally Linear Embedding and Isomap showed a superior performance on all datasets. In very low-dimensional representations and with few differentially expressed genes, these two methods preserve more of the underlying structure of the data than PCA, and thus are favorable alternatives for the visualization of microarray data.

BACKGROUND: Multiple gene expression signatures derived from microarray experiments have been published in the field of leukemia research. A comparison of these signatures with results from new experiments is useful for verification as well as for interpretation of the results obtained. Currently, the percentage of overlapping genes is frequently used to compare published gene signatures against a signature derived from a new experiment. However, it has been shown that the percentage of overlapping genes is of limited use for comparing two experiments due to the variability of gene signatures caused by different array platforms or assay-specific influencing parameters. Here, we present a robust approach for a systematic and quantitative comparison of published gene expression signatures with an exemplary query dataset. RESULTS: A database storing 138 leukemia-related published gene signatures was designed. Each gene signature was manually annotated with terms according to a leukemia-specific taxonomy. Two analysis steps are implemented to compare a new microarray dataset with the results from previous experiments stored and curated in the database. First, the global test method is applied to assess gene signatures and to constitute a ranking among them. In a subsequent analysis step, the focus is shifted from single gene signatures to chromosomal aberrations or molecular mutations as modeled in the taxonomy. Potentially interesting disease characteristics are detected based on the ranking of gene signatures associated with these aberrations stored in the database. Two example analyses are presented. An implementation of the approach is freely available as web-based application. CONCLUSIONS: The presented approach helps researchers to systematically integrate the knowledge derived from numerous microarray experiments into the analysis of a new dataset. By means of example leukemia datasets we demonstrate that this approach detects related experiments as well as related molecular mutations and may help to interpret new microarray data.

The lower the resolution of a given text is, the more dif- ficult it becomes to segment it into single characters. The resolution of screen-rendered text can be very low. This pa- per focuses on smoothed screen-rendered text of very low resolution with typical x-heights of 4 to 7 pixels which is much lower than in other low resolution OCR situations. We propose a recognition-based segmentation algorithm which makes use of oversegmentation by dynamic programming, candidate rating by single character classifiers and a graph based search algorithm for an optimal cut sequence. The al- gorithm is described in detail and experimental results are presented which show the performance on example screen- shot images taken from the public Screen-Word database.

The recognition of screen-rendered text is a novel task. It is performed e.g. by translation tools which allow users to click on any text on the screen and give a translation. Also some commercial OCR programs start to address the problem of reading screenshots. Optical character recognition on screen-shot images can be very challenging due to very small and smoothed fonts. In order to build and compare recognition approaches for screen-rendered text, the availability of standard databases is a fundamental prerequisite. In this paper two freely available databases are presented, one that consists of annotated screenshot images of 28080 single characters and another holding 400 words extracted from documents plus 2 400 generated isolated words. Both databases include meta-information such as x-height, font type, style and rendering conditions. At the example of a developed recognition system, it is shown how these databases can serve for training, testing and optimization.