Overnight accomodations
Visiting Berlin

Computational Diagnostics group

Berlin Center for Genom Based Bioinformatics

International BCB-Workshop on Statistics and Cancer Genomics

August 21, 2003 at 

Magnus-Haus, Am Kupfergraben 7, 10117 Berlin


Michael Boutros

Genome-wide Functional Analysis of Cellular Phenotypes using RNAi

One of the most critical demands following the completion of whole genome sequences is the systematic functional analysis of all predicted gene products. Towards this goal, we established a methodology using RNA-interference (RNAi) in cultured cells to functionally characterize all Drosophila genes required for specific cellular processes. We generated a library of 19,470 double-stranded RNAs (dsRNAs) designed to target by RNAi every predicted open reading frame in the Drosophila genome. High-throughput screens with this dsRNA library comprehensively identified genes required for cell growth and cell survival. Among the 438 targeted genes with the most severe cell viability phenotypes were many without annotated functions. Novel candidate genes were further classified based on the similarity of quantitative phenotypes with known genes required for cell growth and cell survival. In particular, this approach identified a candidate with a further demonstrated anti-apoptotic function encoded by a novel homolog of the mammalian leukemic transcription factors AML1/2. We demonstrate that RNAi screens identify comprehensive sets of functionally related genes that can serve to predict novel gene functions on a genome-wide scale.

Edgar Brunner

Efficient Factorial Designs for cDNA-Microarray Experiments

Based on the Kerr-Churchill ANOVA-model, a linear model for the log-intensities of the expressions in the Cy3- and Cy5-channel is considered. Estimability and variance of the effects of interest as well as the efficiency of the statistics for treatment (variety) effects in factorial designs are discussed. In this context, an average E-optimality of a specific design for a certain set of effects is discussed. As examples, the efficiencies of certain common reference-, swap- , and loop- designs are compared in some one- and two-way layouts. The analysis of the experimental data based on the designs is briefly discussed.

Peter Bühlmann

Supervised Grouping of Genes

A challenging task with expression measurements from thousands of genes is to reveal group of genes whose collective expression is strongly associated with an outcome (response) variable of interest such as an observed tumor type. Unlike unsupervised clustering algorithms, we construct groups of genes by incorporating directly the observed response variable of interest. This is done within the framework of PEnalized LOgistic Regression Analysis and thus, we refer to our method as PELORA. We demonstrate empirically that PELORA identifies relatively few groups of genes whose expression centroids have excellent predictive potential (often superior to state-of-the-art classification methods based on single genes) while still yielding a drastic reduction in dimensionality. Moreover, PELORA can be used in conjunction with additional clinical variables to come up with a model and competitive predictions for medical prognosis. Statistical inference in such a model is proposed by leave-one-out bootstrap methods.

Christine Sers, Oleg Tchernitsa, Ralf-Jürgen Kuban & Reinhold Schäfer

RAS-oncogene dependent gene expression

RAS proteins act as molecular switches transmitting signals from the cell surface to the nucleus via a complex network of kinase cascades. GTP-bound RAS proteins can activate of a number of effector proteins among which Raf, Phosphatidylinositol kinase and RalGDS are the best understood. These effectors stimulate cytoplasmic kinase cascades which result in a profound transcriptional alteration of numerous genes. We have analysed global RAS-dependent transcriptional alterations using suppression subtractive hybridisation for the comparison of immortalized and RAS transformed cells (Zuber et al. 2000). Hundreds of genes encoding proteins with important roles in cell growth control, cytoskeletal organisation and transcriptional regulation are either up or down-regulated in RAS-transformed cells. Further analysis using standard Northern Blots and high density microarrays showed that subsets of genes are regulated by distinct pathways. We were able to define a MAP/ERK module, a PI-3 kinase module and a methylation-dependent module of signal-regulated transcriptional targets. To define the temporal dependency of gene regulation in response to RAS activtion, we used immortal rat fibroblasts harbouring an IPTG-inducible HRAS oncogene. We analysed the expression of 286 selected RAS targets genes at distinct time points after the induction of RAS expression using a cDNA array. We measured RAS protein levels and activation of the MAP/ERK signaling pathway 10, 30 and 60 minutes and 2, 6, 12 , 24, 48 and 72 hours after addition of IPTG. RNA was prepared at these time points and hybridized to the cDNA array. In total, 249 genes were detectable. Among these, several groups were recognized which revealed distinct patterns of up-or down-regulation following the induction of RAS. Thirty-seven genes were found up-regulated and 44 genes were found down-regulated by a factor of 2 or more over time. The most dramatic changes in gene expression were observed within the first 30 minutes after RAS activation and 24 to 48 hours later. For several genes e.g. Fra-1, Srf, MMP3, RhoC and c-Jun such a bisphasic activation was detected. In addition, the inactivation of potential classII tumor suppressor genes (e.g. lox, clusterin) was found to decrease gradually over time. These genes can be re-expression by 5-aza-deoxycytidine in cells stably transformed by oncogenic RAS. This suggests that RAS signalling down-regulates a subset of genes by activation of the methylation machinery.

Terry Speed

Robustness and QC for sets of GeneChips

Affymetrix offers a battery of measures for assessing the quality of a GeneChip experiment, but almost all relate to the experimental process, not directly to the end result, which is a measure of gene expression. Further, they are all single chip measures, with generally recommended ranges of values for the measures; they do not put each chip in the context of other chips in the experiment. By contrast, RMA (robust multi-chip analysis) attempts to do just this, robustly. We find various by-products of the RMA calculation provide valuable quality information. I will outline some of the ways we display and use this information. (This talk describes joint work with Francois Collin, Ben Bolstad and Julia Brettschneider of Berkeley, and Ken Simpson of Melbourne.)

Mike West

Statistical Modelling for Clinico-Genomics: Integrating Metagene Signatures with Clinical Data in Models for Cancer Prognosis

I'll talk about principles, models and practice of clinico-genomics -- the issues, ideas, methods and challenges in the contexts of modelling and intgrating multiple forms of information, including traditional clinical, pathological and demographic factors along with gene expression patterns, to aid in clinical prognosis. Our work in breast cancer provides the contexts, examples, experiences and stimuli for open questions and challenges. One key goal of the talk is to engage clinically-oriented biologists in conversations about embracing rather more complex statistical models than is common in clinical biostatistical practice.

Comments on our webpages please to jaeger@molgen.mpg.de