The Royal Statistical Society

Events Calendar

Search by date or month below. You can also search by region or use our advanced search to search by type of meeting, keywords, speaker, or between specific dates.

Analysis of high-dimensional data in bioinformatics
13 July 2007 10:00 - 17:00 Registration from 09.30
St Leonard''''s Hall, Pollock Hall, Edinburgh



Please use the attached booking form to register

The charges for this event are:
Fellows (Retired/Student) �35 CStats/GradStats �45 Fellows �50 None of the above �65

Bayesian modelling and feature selection of protemic functional data
PHIL BROWN (University of Kent)
We describe the use of Bayesian hierarchical modelling of functional data arising from mass spectroscopy used in proteomics. The data contains both experimental factors and covariates but a desire is to provide interpretation and to discriminate between two or more groups. The functional modelling uses wavelets to accommodate the spikey behaviour. We also look at Bayesian forms of prior distribution which facilitate feature selection in the spirit of the lasso.

Elucidating altered transcriptional programs in breast cancer using Independent Component Analysis
ANDREW TESCHENDORFF (Cancer Research UK)
Most standard methods for analysing tumor-derived gene expression data do not attempt to explicitly infer the altered biological processes underlying cancer. By viewing gene expression as a blind source separation (BSS) problem we can characterise the inferred biological processes in terms of aberrations in functional pathways and transcriptional programs. Using Independent Component Analysis (ICA) to perform BSS, we show that ICA significantly outperforms other linear decomposition techniques. We describe the application of ICA in a meta-analysis of breast cancer, leading to novel associations between biological pathways, regulatory modules and breast cancer phenotypes. Throughout we investigate the role of different ICA contrast functions and novel geometric optimisation methods.

Estimating the number of true discoveries in genome-wide association studies
ARIEF GUSNANTO (MRC Biostatistics Unit, Cambridge)
Recent studies have reported the discovery of more genetic variants of moderate effects associated with disease in genome-wide association studies. This is possible, in spite of the increased number of markers, because the number of samples has also increased. However, the multiplicity problem makes it inevitable for us to get false positive results. A new approach is needed to recognize all potential discoveries in the data while controlling for false discoveries. In this talk, we present a practical model-free procedure to estimate the number of true discoveries as a function of the number of top-ranking SNPs together with the confidence bounds. This approach allows a practical methodology of general utility and produces relevant statistical quantities with simple interpretation. We illustrate our method with applications in 3 studies: (i) prostate cancer study with 6,998 SNPs, (ii) age-related macular degeneration study with 102,249 SNPs and (iii) high-density lipoprotein study with 212,810 SNPs.

Reconstructing gene regulatory networks with Bayesian networks by combining expression data with multiple sources of prior knowledge
DIRK HUSMEIER (BIOSS)
There have been various attempts to reconstruct gene regulatory networks from microarray expression data in the past. However, owing to the limited amount of independent experimental conditions and noise inherent in the measurements, the results have been rather modest so far. For this reason it seems advisable to include biological prior knowledge, related, for instance, to transcription factor binding locations in promoter regions or partially known signalling pathways from the literature. In my presentation, I will describe a Bayesian approach to systematically integrate expression data with multiple sources of prior knowledge. Each source is encoded via a separate energy function, from which a prior distribution over network structures in the form of a Gibbs distribution is constructed. The hyperparameters associated with the different sources of prior knowledge, which measure the influence of the respective prior relative to the data, are sampled from the posterior distribution with MCMC. We have evaluated the proposed scheme on the Raf signalling pathway. Our findings suggest that the inclusion of independent prior knowledge significantly improves the network reconstruction accuracy, and that the values of the hyperparameters inferred with this scheme are close to optimal with respect to minimizing the reconstruction error. This is joint work with Adriano Werhli.

Rapid analysis of pedigree-based genome-wide association studies of quantitative traits
CHRIS HALEY (Roslin Institute, Edinburgh)
The identification of associations between single nucleotide polymorphisms (SNPs) and quantitative trait variation provides a route to the identification of genetic loci controlling such traits in populations of humans and other species. With the possibility of genotyping 106 or more separate SNPs, many traits recorded and the need to account for a potentially complex underlying pedigree, such analyses can be computationally demanding. We report a novel approach for genome-wide pedigree-based QTL association analysis: GRAMMAR (Genome-wide Rapid Association using Mixed Model And Regression). In this method we obtain residuals adjusted for pedigree and other effects from a single mixed-model analysis that does not include SNP genotype. We subsequently analyse the association between these residuals and SNP genotypes using a rapid least squares approach. At the final step, the selected polymorphisms may be followed up with a full mixed-model analysis. Simulation results demonstrate that GRAMMAR is generally similar in power to the full mixed model approach but computationally much more rapid. We illustrate the method with the analysis of data from Genetic Analysis Workshop 15, that involves estimation of associations between 8.5k gene expression traits and 2.7k SNP markers collected from the CEPH human pedigrees. Co-authors: Yurii Aulchenko, Erasmus MC, Rotterdam Alex Lam and DJ de Koning, Roslin Institute

Efficiency of functional regression estimators for combining multiple laser scans of cDNA microarrays
CHRIS GLASBEY (BIOSS)
The first stage in the analysis of cDNA microarray data is estimation of the level of expression of each gene, from laser scans of the hybridised microarrays. Combining multiple laser scans can be formulated as multivariate functional regression through the origin. Maximum likelihood estimation fails, but many alternative estimators exist, one of which is to maximise the likelihood of a Gaussian structural regression model. We show by simulation that, surprisingly, this estimator is efficient in our region of parameter space, even though the distribution of gene expression values is far from Gaussian.

Programme
Full programme including break timings

Documents
753400_Programme 13 July FINAL.doc	753400_Programme 13 July FINAL.doc

Directions to venue:
www.edinburghfirst.com/pdf/Edinburgh%20First%20maps.pdf

Meeting Contact: Christine Hackett [christine@bioss.ac.uk]
Organising Group(s): Bioinformatics Study Group

	July 2
S	M	T	W	T	F	S

31	1	2	3	4	5	6
7	8	9	10	11	12	13
14	15	16	17	18	19	20
21	22	23	24	25	26	27
28	29	30	31	1	2	3

Analysis of high-dimensional data in bioinformatics