Thomas E. Bartlett

A beta-mixture quantile normalization method for correcting probe design bias in Illumina Infinium 450 k DNA methylation data

Motivation: The Illumina Infinium 450 k DNA Methylation Beadchip is a prime candidate technology for Epigenome-Wide Association Studies (EWAS). However, a difficulty associated with these beadarrays is that probes come in two different designs, characterized by widely different DNA methylation distributions and dynamic range, which may bias downstream analyses. A key statistical issue is therefore how best to adjust for the two different probe designs. Results: Here we propose a novel model-based intra-array normalization strategy for 450 k data, called BMIQ (Beta MIxture Quantile dilation), to adjust the beta-values of type2 design probes into a statistical distribution characteristic of type1 probes. The strategy involves application of a three-state beta-mixture model to assign probes to methylation states, subsequent transformation of probabilities into quantiles and finally a methylation-dependent dilation transformation to preserve the monotonicity and continuity of the data. We validate our method on cell-line data, fresh frozen and paraffin-embedded tumour tissue samples and demonstrate that BMIQ compares favourably with two competing methods. Specifically, we show that BMIQ improves the robustness of the normalization procedure, reduces the technical variation and bias of type2 probe values and successfully eliminates the type1 enrichment bias caused by the lower dynamic range of type2 probes. BMIQ will be useful as a preprocessing step for any study using the Illumina Infinium 450 k platform. Availability: BMIQ is freely available from http://code.google.com/p/bmiq/. Contact: a.teschendorff@ucl.ac.uk Supplementary information: Supplementary data are available at Bioinformatics online

Glioblastoma Stem Cells Respond to Differentiation Cues but Fail to Undergo Commitment and Terminal Cell-Cycle Arrest

Summary Glioblastoma (GBM) is an aggressive brain tumor whose growth is driven by stem cell-like cells. BMP signaling triggers cell-cycle exit and differentiation of GBM stem cells (GSCs) and, therefore, might have therapeutic value. However, the epigenetic mechanisms that accompany differentiation remain poorly defined. It is also unclear whether cell-cycle arrest is terminal. Here we find only a subset of GSC cultures exhibit astrocyte differentiation in response to BMP. Although overtly differentiated non-cycling astrocytes are generated, they remain vulnerable to cell-cycle re-entry and fail to appropriately reconfigure DNA methylation patterns. Chromatin accessibility mapping identified loci that failed to alter in response to BMP and these were enriched in SOX transcription factor-binding motifs. SOX transcription factors, therefore, may limit differentiation commitment. A similar propensity for cell-cycle re-entry and de-differentiation was observed in GSC-derived oligodendrocyte-like cells. These findings highlight significant obstacles to BMP-induced differentiation as therapy for GBM.

Corruption of the Intra-Gene DNA Methylation Architecture Is a Hallmark of Cancer

Epigenetic processes - including DNA methylation - are increasingly seen as having a fundamental role in chronic diseases like cancer. It is well known that methylation levels at particular genes or loci differ between normal and diseased tissue. Here we investigate whether the intra-gene methylation architecture is corrupted in cancer and whether the variability of levels of methylation of individual CpGs within a defined gene is able to discriminate cancerous from normal tissue, and is associated with heterogeneous tumour phenotype, as defined by gene expression. We analysed 270985 CpGs annotated to 18272 genes, in 3284 cancerous and 681 normal samples, corresponding to 14 different cancer types. In doing so, we found novel differences in intra-gene methylation pattern across phenotypes, particularly in those genes which are crucial for stem cell biology; our measures of intra-gene methylation architecture are a better determinant of phenotype than measures based on mean methylation level alone (K-S test in all 14 diseases tested). These per-gene methylation measures also represent a considerable reduction in complexity, compared to conventional per-CpG beta-values. Our findings strongly support the view that intra-gene methylation architecture has great clinical potential for the development of DNA-based cancer biomarkers.

A DNA Methylation Network Interaction Measure, and Detection of Network Oncomarkers

Epigenetic processes–including DNA methylation–are increasingly seen as having a fundamental role in chronic diseases like cancer. DNA methylation patterns offer a route to develop prognostic measures based directly on DNA measurements, rather than less-stable RNA measurements. A novel DNA methylation-based measure of the co-ordinated interactive behaviour of genes is developed, in a network context. It is shown that this measure reflects well the co-regulatory behaviour linked to gene expression (at the mRNA level) over the same network interactions. This measure, defined for pairs of genes in a single patient/sample, associates with overall survival outcome independent of known prognostic clinical features, in several independent data sets relating to different cancer types. In total, more than half a billion CpGs in over 1600 samples, taken from nine different cancer entities, are analysed. It is found that groups of gene-pair interactions which associate significantly with survival identify statistically significant subnetwork modules. Many of these subnetwork modules are shown to be biologically relevant by strong correlation with pre-defined gene sets, such as immune function, wound healing, mitochondrial function and MAP-kinase signalling. In particular, the wound healing module corresponds to an increase in co-ordinated interactive behaviour between genes for worse prognosis, and the immune module corresponds to a decrease in co-ordinated interactive behaviour between genes for worse prognosis. This measure has great potential for defining DNA-based cancer biomarkers. Such biomarkers could naturally be developed further, by drawing on the rapidly expanding knowledge base of network science.

Network inference and community detection, based on covariance matrices, correlations, and test statistics from arbitrary distributions

ABSTRACT In this article we propose methodology for inference of binary-valued adjacency matrices from various measures of the strength of association between pairs of network nodes, or more generally pairs of variables. This strength of association can be quantified by sample covariance and correlation matrices, and more generally by test-statistics and hypothesis test p-values from arbitrary distributions. Community detection methods such as block modeling typically require binary-valued adjacency matrices as a starting point. Hence, a main motivation for the methodology we propose is to obtain binary-valued adjacency matrices from such pairwise measures of strength of association between variables. The proposed methodology is applicable to large high-dimensional data sets and is based on computationally efficient algorithms. We illustrate its utility in a range of contexts and data sets.

Intra-Gene DNA Methylation Variability Is a Clinically Independent Prognostic Marker in Women's Cancers.

We introduce a novel per-gene measure of intra-gene DNA methylation variability (IGV) based on the Illumina Infinium HumanMethylation450 platform, which is prognostic independently of well-known predictors of clinical outcome. Using IGV, we derive a robust gene-panel prognostic signature for ovarian cancer (OC, n = 221), which validates in two independent data sets from Mayo Clinic (n = 198) and TCGA (n = 358), with significance of p = 0.004 in both sets. The OC prognostic signature gene-panel is comprised of four gene groups, which represent distinct biological processes. We show the IGV measurements of these gene groups are most likely a reflection of a mixture of intra-tumour heterogeneity and transcription factor (TF) binding/activity. IGV can be used to predict clinical outcome in patients individually, providing a surrogate read-out of hard-to-measure disease processes.

Single-cell Co-expression Subnetwork Analysis

Single-cell transcriptomic data have rapidly become very popular in genomic science. Genomic science also has a long history of using network models to understand the way in which genes work together to carry out specific biological functions. However, working with single-cell data presents major challenges, such as zero inflation and technical noise. These challenges require methods to be specifically adapted to the context of single-cell data. Recently, much effort has been made to develop the theory behind statistical network models. This has lead to many new models being proposed, and has provided a thorough understanding of the properties of existing models. However, a large amount of this work assumes binary-valued relationships between network nodes, whereas genomic network analysis is traditionally based on continuous-valued correlations between genes. In this paper, we assess several established methods for genomic network analysis, we compare ways that these methods can be adapted to the single-cell context, and we use mixture-models to infer binary-valued relationships based on gene-gene correlations. Based on these binary relationships, we find that excellent results can be achieved by using subnetwork analysis methodology popular amongst network statisticians. This methodology thereby allows detection of functional subnetwork modules within these single-cell genomic networks.

A Power Variance Test for Nonstationarity in Complex-Valued Signals

We propose a novel algorithm for testing the hypothesis of nonstationarity in complex-valued signals. The implementation uses both the bootstrap and the Fast Fourier Transform such that the algorithm can be efficiently implemented in O(NlogN) time, where N is the length of the observed signal. The test procedure examines the second-order structure and contrasts the observed power variance -- i.e. the variability of the instantaneous variance over time -- with the expected characteristics of stationary signals generated via the bootstrap method. Our algorithmic procedure is capable of learning different types of nonstationarity, such as jumps or strong sinusoidal components. We illustrate the utility of our test and algorithm through application to turbulent flow data from fluid dynamics.