Alicia Oshlack

A scaling normalization method for differential expression analysis of RNA-seq data

The fine detail provided by sequencing-based transcriptome surveys suggests that RNA-seq is likely to become the platform of choice for interrogating steady state RNA. In order to discover biologically important changes in expression, we show that normalization continues to be an essential step in the analysis. We outline a simple and effective method for performing normalization and show dramatically improved results for inferring differential expression in simulated and publicly available data sets.

Gene ontology analysis for RNA-seq: accounting for selection bias

We present GOseq, an application for performing Gene Ontology (GO) analysis on RNA-seq data. GO analysis is widely used to reduce complexity and highlight biological processes in genome-wide expression studies, but standard methods give biased results on RNA-seq data due to over-detection of differential expression for long and highly expressed transcripts. Application of GOseq to a prostate cancer data set shows that GOseq dramatically changes the results, highlighting categories more consistent with the known biology.

A comparison of background correction methods for two-colour microarrays

MOTIVATION Microarray data must be background corrected to remove the effects of non-specific binding or spatial heterogeneity across the array, but this practice typically causes other problems such as negative corrected intensities and high variability of low intensity log-ratios. Different estimators of background, and various model-based processing methods, are compared in this study in search of the best option for differential expression analyses of small microarray experiments. RESULTS Using data where some independent truth in gene expression is known, eight different background correction alternatives are compared, in terms of precision and bias of the resulting gene expression measures, and in terms of their ability to detect differentially expressed genes as judged by two popular algorithms, SAM and limma eBayes. A new background processing method (normexp) is introduced which is based on a convolution model. The model-based correction methods are shown to be markedly superior to the usual practice of subtracting local background estimates. Methods which stabilize the variances of the log-ratios along the intensity range perform the best. The normexp+offset method is found to give the lowest false discovery rate overall, followed by morph and vsn. Like vsn, normexp is applicable to most types of two-colour microarray data. AVAILABILITY The background correction methods compared in this article are available in the R package limma (Smyth, 2005) from http://www.bioconductor.org. SUPPLEMENTARY INFORMATION Supplementary data are available from http://bioinf.wehi.edu.au/resources/webReferences.html.

From RNA-seq reads to differential expression results

Many methods and tools are available for preprocessing high-throughput RNA sequencing data and detecting differential expression.

SWAN: Subset-quantile Within Array Normalization for Illumina Infinium HumanMethylation450 BeadChips

DNA methylation is the most widely studied epigenetic mark and is known to be essential to normal development and frequently disrupted in disease. The Illumina HumanMethylation450 BeadChip assays the methylation status of CpGs at 485,577 sites across the genome. Here we present Subset-quantile Within Array Normalization (SWAN), a new method that substantially improves the results from this platform by reducing technical variation within and between arrays. SWAN is available in the minfi Bioconductor package.

Transcript length bias in RNA-seq data confounds systems biology

BackgroundSeveral recent studies have demonstrated the effectiveness of deep sequencing for transcriptome analysis (RNA-seq) in mammals. As RNA-seq becomes more affordable, whole genome transcriptional profiling is likely to become the platform of choice for species with good genomic sequences. As yet, a rigorous analysis methodology has not been developed and we are still in the stages of exploring the features of the data.ResultsWe investigated the effect of transcript length bias in RNA-seq data using three different published data sets. For standard analyses using aggregated tag counts for each gene, the ability to call differentially expressed genes between samples is strongly associated with the length of the transcript.ConclusionTranscript length bias for calling differentially expressed genes is a general feature of current protocols for RNA-seq technology. This has implications for the ranking of differentially expressed genes, and in particular may introduce bias in gene set testing for pathway analysis and other multi-gene systems biology analyses.ReviewersThis article was reviewed by Rohan Williams (nominated by Gavin Huttley), Nicole Cloonan (nominated by Mark Ragan) and James Bullard (nominated by Sandrine Dudoit).

Expression profiling in primates reveals a rapid evolution of human transcription factors.

Although it has been hypothesized for thirty years that many human adaptations are likely to be due to changes in gene regulation, almost nothing is known about the modes of natural selection acting on regulation in primates. Here we identify a set of genes for which expression is evolving under natural selection. We use a new multi-species complementary DNA array to compare steady-state messenger RNA levels in liver tissues within and between humans, chimpanzees, orangutans and rhesus macaques. Using estimates from a linear mixed model, we identify a set of genes for which expression levels have remained constant across the entire phylogeny (∼70 million years), and are therefore likely to be under stabilizing selection. Among the top candidates are five genes with expression levels that have previously been shown to be altered in liver carcinoma. We also find a number of genes with similar expression levels among non-human primates but significantly elevated or reduced expression in the human lineage, features that point to the action of directional selection. Among the gene set with a human-specific increase in expression, there is an excess of transcription factors; the same is not true for genes with increased expression in chimpanzee.

Array-Based Gene Discovery with Three Unrelated Subjects Shows SCARB2/LIMP-2 Deficiency Causes Myoclonus Epilepsy and Glomerulosclerosis

Action myoclonus-renal failure syndrome (AMRF) is an autosomal-recessive disorder with the remarkable combination of focal glomerulosclerosis, frequently with glomerular collapse, and progressive myoclonus epilepsy associated with storage material in the brain. Here, we employed a novel combination of molecular strategies to find the responsible gene and show its effects in an animal model. Utilizing only three unrelated affected individuals and their relatives, we used homozygosity mapping with single-nucleotide polymorphism chips to localize AMRF. We then used microarray-expression analysis to prioritize candidates prior to sequencing. The disorder was mapped to 4q13-21, and microarray-expression analysis identified SCARB2/Limp2, which encodes a lysosomal-membrane protein, as the likely candidate. Mutations in SCARB2/Limp2 were found in all three families used for mapping and subsequently confirmed in two other unrelated AMRF families. The mutations were associated with lack of SCARB2 protein. Reanalysis of an existing Limp2 knockout mouse showed intracellular inclusions in cerebral and cerebellar cortex, and the kidneys showed subtle glomerular changes. This study highlights that recessive genes can be identified with a very small number of subjects. The ancestral lysosomal-membrane protein SCARB2/LIMP-2 is responsible for AMRF. The heterogeneous pathology in the kidney and brain suggests that SCARB2/Limp2 has pleiotropic effects that may be relevant to understanding the pathogenesis of other forms of glomerulosclerosis or collapse and myoclonic epilepsies.

ChIP-seq analysis reveals distinct H3K27me3 profiles that correlate with transcriptional activity

Transcriptional control is dependent on a vast network of epigenetic modifications. One epigenetic mark of particular interest is tri-methylation of lysine 27 on histone H3 (H3K27me3), which is catalysed and maintained by Polycomb Repressive Complex 2 (PRC2). Although this histone mark is studied widely, the precise relationship between its local pattern of enrichment and regulation of gene expression is currently unclear. We have used ChIP-seq to generate genome-wide maps of H3K27me3 enrichment, and have identified three enrichment profiles with distinct regulatory consequences. First, a broad domain of H3K27me3 enrichment across the body of genes corresponds to the canonical view of H3K27me3 as inhibitory to transcription. Second, a peak of enrichment around the transcription start site (TSS) is commonly associated with ‘bivalent’ genes, where H3K4me3 also marks the TSS. Finally and most surprisingly, we identified an enrichment profile with a peak in the promoter of genes that is associated with active transcription. Genes with each of these three profiles were found in different proportions in each of the cell types studied. The data analysis techniques developed here will be useful for the identification of common enrichment profiles for other histone modifications that have important consequences for transcriptional regulation.

Optimizing the noise versus bias trade-off for Illumina whole genome expression BeadChips

Five strategies for pre-processing intensities from Illumina expression BeadChips are assessed from the point of view of precision and bias. The strategies include a popular variance stabilizing transformation and model-based background corrections that either use or ignore the control probes. Four calibration data sets are used to evaluate precision, bias and false discovery rate (FDR). The original algorithms are shown to have operating characteristics that are not easily comparable. Some tend to minimize noise while others minimize bias. Each original algorithm is shown to have an innate intensity offset, by which unlogged intensities are bounded away from zero, and the size of this offset determines its position on the noise–bias spectrum. By adding extra offsets, a continuum of related algorithms with different noise–bias trade-offs is generated, allowing direct comparison of the performance of the strategies on equivalent terms. Adding a positive offset is shown to decrease the FDR of each original algorithm. The potential of each strategy to generate an algorithm with an optimal noise–bias trade-off is explored by finding the offset that minimizes its FDR. The use of control probes as part of the background correction and normalization strategy is shown to achieve the lowest FDR for a given bias.