UCL Cancer Institute

Statistical Cancer Genomics

Group Leader: Dr Andrew Teschendorff

Statistical Cancer Genomics

We develop and apply advanced statistical methodology to enable a more meaningful interpretation of large scale multi-dimensional cancer genomic data. Specifically, we are applying tools from network theory, Bayesian statistics, signal processing and machine learning to help address some of the outstanding challenges in the cancer genomics and epigenomics fields.


Cancer system-omics

Recent experimental and theoretical work have shown that differential networks are key to understanding the rewiring of the cellular network under endogenous and exogenous perturbations, including those characterizing cancer. We are thus exploring novel tools and concepts from graph theory and network physics to help us elucidate the system-omic principles underlying the cancer phenotype. For example, we recently showed that metastatic cancers are characterised by an increased dynamical network entropy and that changes in dynamical entropy can be used to pinpoint nodes in the network where rewiring takes place (see Teschendorff AE and Severini S, BMC Systems Biology 2010 4:104).

We are also interested in graph theoretical methods for analyzing interaction networks integrated with multi-dimensional cancer genomic data (mRNA expression, copy number and DNA methylation). The specific goals are to:

(i) Identify the genomic/epigenomic aberrations and associated signaling pathways driving the carcinogenic process.

(ii) To explore the statistical properties of cancer gene networks and investigate the relationship of these network properties with cancer phenotypes.

Members and collaborators with an interest in this area: Andrew Teschendorff, James West, Simone Severini (Computer Science/Physics-UCL), Ginestra Bianconi (Physics, Boston USA), Lucas Lacasa (Applied Maths, Madrid).

Statistical analysis of large scale DNA methylation data

There is a rapidly growing interest in using epigenetic markers for the early detection, diagnosis and prognosis of cancers. DNA methylation profiling using Illumina Infinium beadarrays offers the scalability and coverage for large scale studies that aim to identify such markers. However, identification of such markers can be difficult due to small effect sizes and numerous confounding factors including beadchip effects. Our aim is therefore to develop novel statistical methods to enable a more meaningful interpretation of such data. Specifically,

(i) We have recently developed a signal processing algorithm based on Independent Component Analysis, called Independent Surrogate Variable Analysis (ISVA), to improve feature selection in the presence of potential confounding factors (see Teschendorff AE et al Bioinformatics 2011, 27(11)). This tool is freely available as an R-package (isva) from CRAN (www.r-project.org).

(ii) We are currently exploring signal processing tools used in time-series data for the analysis of spatially correlated DNA methylation patterns to obtain global markers of early detection, diagnosis or prognosis of cancers.

(iii) We are currently investigating variational Bayesian beta mixture models for feature selection and clustering problems in DNA methylation data.

(iv) We are also currently developing improved normalization methods for Illumina Infinium 450k technology.

Members and collaborators with an interest in this area: Andrew Teschendorff, Joanna Zhuang, James West, Tom Bartlett, Zhanyu Ma (Electrical Engineering, Stockholm), Martin Widschwendter (IfWH-UCL), Stephan Beck (Medical Genomics Group), Alexey Zaikin (Maths-UCL), Sofia Olhede (Statistics-UCL), Reimer Kuehn and Peter Sollich (Maths-KCL).

Cancer Epigenomics

We are currently involved in a large number of data-driven projects in cancer epigenomics in close collaboration with biologists and clinicians at UCL and elsewhere, in which we provide statistical support. These include:

(i) Epigenetic stem cell models of cancer: we previously showed that genes, which are required for differentiation of stem cells and which exhibit widespread hypermethylation in cancer, also exhibit age-associated hypermethylation in normal tissue (Teschendorff AE et al Genome Research 2010, 20(4)). Currently, we are exploring if this age associated DNA methylation signature can indicate the risk of neoplastic transformation in a number of prospective studies.

(ii) Epigenomics of head & neck cancers: we are currently investigating DNA methylation profiles of head & neck cancers in order to identify novel molecular subclasses of clinical relevance.

(iii) Meta-analyses: we are currently using meta-analysis approaches to identify robust and tissue independent prognostic DNA methylation signatures.

Members and collaborators with an interest in this area: Andrew Teschendorff, Joanna Zhuang, Martin Widschwendter (IfWH-UCL), Matthias Lechner & Stephan Beck (Medical Genomics Group), Chris Boshoff (Viral Oncology Group).

Molecular classification of breast cancer

Generating maps of signaling pathway activity is an important goal in cancer genomics, as these maps can provide a clinically more meaningful molecular taxonomy of heterogeneous cancers. In-vitro derived perturbation signatures that reflect the transcriptomic changes of gene perturbations in model systems or more generally prior pathway models can be used in combination with large-scale mRNA expression data to generate maps of pathway activity and thus the potential to guide therapies. To this end, we recently developed an algorithm called DART (Denoising algorithm based on Relevance network Topology) (see Jiao et al. BMC Bioinformatics 2011 12:403) which improves the inference of molecular pathway activity levels in primary tumours. A similar algorithm was also used in our previous work (see Teschendorff AE et al. BMC Cancer 2010 10:604) to dissect bulk breast cancer profiles in terms of stromal immune cell signatures. This showed that differential Th1/Th2 type immune responses and TGFbeta pathway activity can provide improved prognostic stratifications of hormone receptor negative breast cancer.

Currently, we are exploring these denoising tools in the context of transcription factor ChIP-Seq binding data to help identify novel molecular subclasses of breast cancer.

Members and collaborators with an interest in this area: Andrew Teschendorff, Nicky McGranahan (LRI, London), Carlos Caldas & Jason Carroll (CRI, Cambridge).



Selected Publications

Zhuang JJ, Widschwendter M, Teschendorff AE. A comparison of feature selection and classification methods in DNA methylation studies using the Illumina Infinium platform. BMC Bioinformatics. 2012 Apr 24;13(1):59. PubMed

Teschendorff AE, Widschwendter M. Differential variability improves the identification of cancer risk markers in DNA methylation studies profiling precursor cancer lesions. Bioinformatics. 2012 Apr 6. PubMed

Teschendorff AE, Jones A, Fiegl H, Sargent A, Zhuang JJ, Kitchener HC, Widschwendter M. Epigenetic variability in cells of normal cytology is associated with the risk of future morphological transformation. Genome Med. 2012 Mar 27;4(3):24. PubMed

Zhuang J, Jones A, Lee SH, Ng E, Fiegl H, Zikan M, Cibula D, Sargent A, Salvesen HB, Jacobs IJ, Kitchener HC, Teschendorff AE, Widschwendter M. The dynamics and prognostic potential of DNA methylation changes at stem cell gene loci in women's cancer. PLoS Genet. 2012 Feb;8(2):e1002517. Epub 2012 Feb 9. PubMed

Ross-Innes CS, Stark R, Teschendorff AE, Holmes KA, Ali HR, Dunning MJ, Brown GD, Gojis O, Ellis IO, Green AR, Ali S, Chin SF, Palmieri C, Caldas C, Carroll JS. Differential oestrogen receptor binding is associated with clinical outcome in breast cancer. Nature. 2012 Jan 4;481(7381):389-93. PubMed

Jiao Y, Lawler K, Patel GS, Purushotham A, Jones AF, Grigoriadis A, Tutt A, Ng T, Teschendorff AE. DART: Denoising Algorithm based on Relevance network Topology improves molecular pathway activity inference. BMC Bioinformatics. 2011 Oct 19;12(1):403. Pubmed

Teschendorff AE, Zhuang J, Widschwendter M. Independent surrogate variable analysis to deconvolve confounding factors in large-scale microarray profiling studies. Bioinformatics. 2011 Jun 1;27(11):1496-505. Epub 2011 Apr 6. Pubmed

Feber A, Wilson GA, Zhang L, Presneau N, Idowu B, Down TA, Rakyan VK, Noon LA, Lloyd AC, Stupka E, Schiza V, Teschendorff AE, Schroth GP, Flanagan A, Beck S. Comparative methylome analysis of benign and malignant peripheral nerve sheath tumors. Genome Res. 2011 Apr;21(4):515-24. Epub 2011 Feb 1. Pubmed

Teschendorff AE, Jiao Y, Caldas C. Prognostic gene network modules in breast cancer hold promise. Breast Cancer Res. 2010 Dec 8;12(6):317. Pubmed

Teschendorff AE, Gomez S, Arenas A, El-Ashry D, Schmidt M, Gehrmann M, Caldas C. Improved prognostic classification of breast cancer defined by antagonistic activation patterns of immune response pathway modules. BMC Cancer. 2010 Nov 4;10:604. Pubmed

Bell CG, Teschendorff AE, Rakyan VK, Maxwell AP, Beck S, Savage DA. Genome-wide DNA methylation analysis for diabetic nephropathy in type 1 diabetes mellitus. BMC Med Genomics. 2010 Aug 5;3:33. Pubmed

More publications, click here.



Group Leader

Andrew Teschendorff

Andrew Teschendorff, PhD
Heller Research Fellow
UCL Cancer Institute
University College London
Paul O’Gorman Building
72 Huntley Street
London WC1E 6BT, UK
Tel: +44-20-7679-0727


James West (PhD student) (CoMPLEX, co-supervised with Dr Simone Severini (Physics))

Chris Banerji

Chris Banerji (PhD student) (CoMPLEX, co-supervised with Dr. Simone Severini)

Shazia Anjum PhD (Research Associate, co-supervised with Prof Martin Widschwendter)

Former Members

Joanna Zhuang PhD: 2010-2012, Postdoc (CBRC, co-supervised with Prof Martin Widschwendter (IfWH))

Yan Jiao PhD: 2010-2011, Postdoc (CCIC, co-supervised with Prof Tony Ng (KCL))


Research supported by

Michael & Morven Heller (Heller Research Fellowship)