The scientific program will be composed of several sessions of invited and contributed talks. Also, there will be poster sessions. The schedule can be downloaded from here .

The complete workshop proceedings can be downloaded from here .

Florence d'Alché Buc, Université d'Evry-Val d'Essonne, Evry, France.

Protein-protein network inference with regularized output and input kernel methods

Prediction of a physical interaction between two proteins has been addressed in the context of supervised learning, unsupervised learning and more recently, semi-supervised learning using various sources of information (genomic, phylogenetic, protein localization and function). The problem can be seen as a kernel matrix completion task if one defines a kernel that encodes similarity between proteins as nodes in a graph or alternatively, as a binary supervised classification task where inputs are pairs of proteins.
In this talk, we first make a review of existing works (matrix completion, SVM for pairs, metric learning, training set expansion), identifying the relevant features of each approach. Then we define the framework of output kernel regression (OKR) that uses the kernel trick in the output feature space. After recalling the results obtained so far with tree-based output kernel regression methods, we develop a new family of methods based on Kernel Ridge Regression that benefit from the use of kernels both in the input feature space and the output feature space. The main interest of such methods is that imposing various regularization constraints still leads to closed form solutions. We show especially how such an approach allows to handle unlabeled data in a transductive setting of the network inference problem and multiple networks in a multi-task like inference problem.
New results on simulated data and yeast data illustrate the talk.


Nir Friedman, The Hebrew University of Jerusalem, Jerusalem, Israel.

Exploring transcription regulation through cell-to-cell variability

The regulation of cellular protein levels is a complex process involving many regulatory mechanisms. These regulatory mechanisms introduce a cascade of stochastic events leading to variability of protein levels between cells. Previous studies have shown that perturbing genes involved in transcription regulation alters variability of protein levels, but to date, there has been no systematic characterization of these effects. Here we utilize single-cell expression levels of two fluorescent reporters under a wide range of genetic perturbations in Saccharomyces cerevisiae to identify proteins that affect expression variability.
We introduce computational methodology to determine the variability introduced by each perturbation, and distinguish between global variability, affecting both reporters in a coordinated manner, and local variability, affecting individual reporters independently. Classifying genes by their variability phenotype identifies functionally coherent groups, which broadly correlate with the different stages of transcriptional regulation. Specifically, we find that perturbation of processes related to DNA maintenance, chromatin regulation and RNA synthesis affect local variability, while processes related to protein synthesis and transport, cell morphology and cell size affect global variability. In addition, we find that perturbations of many processes related to chromatin regulation affect both global and local variability. Finally, we demonstrate that the variability phenotypes of different protein complexes provide insights into their cellular functions. Our methodology provides tools for examining arising data on variability, and establishes the utility of this phenotype as a tool in dissecting the regulatory mechanisms involved in gene expression.


Ursula Kummer, BIOQUANT, University of Heidelberg, Germany.

Computational environments for modeling biochemical networks

Computational modeling is an integral and crucial part of systems biology. It relies on accessible and user-friendly software to set up models, model management and model analysis. Here, two systems are presented that have been implemented for these needs. The first one, COPASI, has been around since 2004 and is a standalone software suite that encompasses many of the commonly used algorithms and approaches in computational modeling. Amongst others, it allows parameter estimation of model on the basis of experimental data sets with diverse methods. The second software is SYCAMORE which is a web based application designed to allow database driven modeling. Thus, it interacts directly with databases for enzymatic kinetics and with tools to estimate parameters based on protein structural data. Both systems are constantly refined and features added.


Hans Lehrach, Max Planck Institute for Molecular Genetics, Berlin, Germany.

Deep sequencing and systems biology: steps on the way to an individualised treatment of cancer patients

Biological processes are driven by complex networks of interactions between molecular and cellular components. Predicting the outcome of potential disturbances is of prime importance to be able to prevent disease, as well as to identify possible therapies for diseases, which are already present. To predict the behaviour of such complex networks, we will have to develop general models of the processes involved, based on information on pathways derived from genetic and molecular approaches, to ‘individualise’ these by applying ‘genomics’ scale analysis techniques (e.g. genome and/or transcriptome analysis by next-gen sequencing techniques-genomics), and to explore the behaviour of these models computationally (systems biology). We are using a combination of high throughput sequencing of genome and transcriptome of both tumor and patient to establish predictive models (virtual patients), which ultimately will reflect the response of real patients to specific therapies in oncology and other areas of medicine.


Vebjorn Ljosa, The Broad Institute of MIT and Harvard, USA.

Automatic quantification of subtle cellular phenotypes in microscopy-based high-throughput experiments

Microscopy-based high-throughput experiments can provide a view into biological responses and states at the resolution of singe cells. CellProfiler, our open-source image-analysis software, has become widely used by biologists to design custom analysis pipelines for complex high-throughput assays. I will discuss our work in progress to automatically quantify the prevalence of subtle cellular phenotypes in high-throughput samples of cultured cells I will also touch briedly on the use of machine learning to improve the accuracy and robustness of CellProfiler's image segmentation.
Our classification tool, CellProfiler Analyst, enables a biologist to train a boosting classifier iteratively to detect rare, complex phenotypes, and its usefulness has been demonstrated in several high-throughput screens. Here, I will describe a method to learn phenotypes without requiring hand-labeled cells for training. Instead, a classifier is trained from negative and positive controls in the experiment, where the positives are known to be enriched in the phenotype of interest, even if only slightly (e.g., 55% vs. 45% penetrance). By nonlinearly projecting cells into a random feature space, we can use efficient linear methods but still benefit from nonlinear notions of similarity, and can overcome experimental noise by training on millions of cells. Using the resulting classifier to assign soft labels to each cell in the experiment, we can identify enriched samples ("hits") nonparametrically. Furthermore, we are developing techniques to automatically identify relevant cellular phenotypes in large-scale chemical profiling experiments.

Oral presentations