computational
modeling in biology

Upcoming talks

18.04.2013  - Bettina Knapp

09:15 a.m., room 160 a+b, building 56

• Title 
  Perturbation data: Statistical Analysis, Phenotypic Effects, Network Inference
• Abstract 
  Perturbation experiments such as RNA interference (RNAi) allow the simultaneous screening of hundreds to thousands of genes in a high-content manner. A detailed quantification of perturbation effects on specific phenotypes can be assessed using multiparametric imaging. This allows to identify genes which are involved in certain biological processes and to elucidate the gene function. We present an approach which uses individual cell measurements for the data analysis and statistical scoring of RNAi data. We show that the phenotypic effect after a perturbation is highly influenced by each cell's population context. Taking therefore the single cell information into account during data analysis, an increased sensitivity and specificity in comparison to already existing methods can be observed. Having identified significant factors of a biological process, the spatial and temporal placement of these factors in the underlying networks is still a challenging task. We introduce a network inference approach for perturbation data which is based on a linear program and thus, can be solved efficiently even for large-scale problems. Based on simulated data we show an improved performance of our approach over state-of-the art methods. Using our approach on real data to study the intracellular signaling of human primary naive CD4+ T-cells as well as ErbB signaling in trastuzumab resistant breast cancer cells we can recover already known interactions and additionally new ones. The interactions inferred with our approach for the ErbB data predict an important role of negative and positive feedback in controlling the cell cycle progression which needs to be validated further. In conclusion, the use of appropriate methods to analyze perturbation data and to infer the network topology allows to optimally exploit the data and to address the biological question at hand.

Past talks

14.03.2013  - Julien Gagneur

09:15 a.m., room 160 a+b, building 56

• Title 
  Systems genetics with multiple environments
• Abstract 
  Dissecting the molecular mechanisms that link genotype to phenotype promises to deliver the necessary insights to develop drugs tailored to the genetic background and life circumstances of the patient. Information from interventional data is scarce, and hence the challenge resides in developing causal inference strategies to exploit the breadth of observational population-level genetic and molecular profiling data being generated.
  
  Here we investigated to what extent environmental perturbations, combined with genetic variations, facilitate causal inference in molecular networks. Using yeast as a model system, we carried out joint profiling of fitness and gene expression of a genetically diverse population in 5 environmental contexts. We developed novel inference techniques to predict molecular functional intermediates with an environment-specific role on growth. Our approaches leverages on ubiquitous genotype-environment interactions, exploiting the rich statistical independencies they imply. Technically, we build on Bayesian model comparisons, assessing the statistical evidence that a particular transcript carries a mediating role between genetic signal and its environment-specific effect on phenotype. We applied the approach to genome-wide identified transcripts specific for each environment-specific growth QTLs. Comprehensive independent test using the genome-wide deletion collection confirmed the majority of the 400 top-ranking model predictions. Our results show that exploiting condition-specific genetic effects substantially increases the predictive accuracy over approaches based on genetic or environmental variations alone.
  
  Together, these results have wide-ranging implications for the design of clinical omics studies and their integrated analysis across multiple contexts.

13.03.2013  - Andrea Ocone

10 a.m., room 160 a+b, building 56

• Title 
  Hybrid stochastic models: a statistically tractable approach to model regulatory network dynamics
• Abstract 
  Computational modelling of the dynamics of gene regulatory networks is a central task of systems biology. For networks of small/medium scale, the dominant paradigm is represented by systems of coupled nonlinear ordinary differential equations (ODEs). ODEs afford great mechanistic detail and flexibility, but calibrating these models to data is often an extremely difficult statistical problem. I will present a general statistical inference framework for stochastic transcription-translation networks. The model is based on a hybrid representation of the system, with (binary) promoter and (continuous) protein variables. Inference can be carried out with an efficient variational approach which is computationally scalable with the number of genes in the network. As an application, I will show how this approach can be used to model the circadian clock of the picoalga Ostreococcus tauri.

11.03.2013  - Matthias Reumann

10 a.m., room 160 a+b, building 56

• Title 
  High performance computing enabling exhaustive analysis of higher order single nucleotide polymorphism interaction in Genome Wide Association Studies
• Abstract 
  Genome-wide association studies (GWAS) have become a common platform for systematic discovery of loci associated with disease. Univariate approaches commonly employed may miss important SNP associations that only appear through multivariate analysis in complex diseases. However, the latter are currently limited by its inherent computational complexity. We present a computational framework that harnesses supercomputers. Based on our results , we estimate a three-way interaction analysis on 1.1M SNP GWAS data requiring over 8.5 years on the full “Avoca” IBM Blue Gene/Q installation at the Victorian Life Sciences Computation Initiative. However, on the currently largest IBM Blue Gene/Q installation “Sequoia” at Lawrence Livermore National Laboratory this could take only about a month given linear scaling. Thus, our framework makes possible exhaustive analysis of higher order interaction studies and enables previously infeasible GWAS analysis techniques.

28.11.2012  - Dana Lahat

1 pm, room 160 a+b, building 56

• Title
  Second-Order Multidimensional ICA: Theory and Methods
• Abstract
  Independent component analysis (ICA) and blind source separation (BSS) deal with extracting a number of mutually independent elements from a set of observed linear mixtures. Motivated by various applications, this talk considers a more general and more flexible model: the sources can be partitioned into groups exhibiting dependence within a given group but independence between two different groups.
   
  The core of this work is the statistical analysis of the blind separation of multidimensional components based on second-order statistics, in a piecewise-stationary model. We develop the likelihood and the associated estimating equations for the Gaussian case. We obtain closed-form expressions for the Fisher information matrix and the Cramér-Rao lower bound (CRLB) of the de-mixing parameters, as well as the mean square error (MSE) of the component estimates. For Gaussian data, our separation criterion achieves, up to higher-order terms, the CRLB, and is thus optimal in the MSE sense.
   
  We then turn to the case when the separation procedure is based on a one-dimensional model, followed by a clustering step, in which the one-dimensional output is assigned into groups, representing the multidimensional components. We show that for piecewise stationary data, and when only second-order statistics are used, this form of separation is suboptimal.
   
  We demonstrate our methods and algorithms on an astrophysical application. Namely, the extraction of the Cosmic Microwave Background Radiation from its observations.

21.06.2012 - Atefeh Kazeroonian

• Title
  Fast Fourier Transform and Fast Multipole Methods for efficient N-particle simulation problems
• Abstract
  The problem of analyzing a system of N bodies with pairwise interactions emerges in many classic and modern scientic elds. Examples include N-body systems with Coulomb or gravitational forces, motion of molecules in a fluid, magnetic spins in a ferromagnetic substance, etc. In this sort of problems, a simulation of the macroscopic behavior of the system requires all the pairwise interactions among the system components to be taken into account, which leads to O(N2) computations.
  The particular system under study in this thesis is a lattice Ising-like spin system with periodic boundary conditions, where spins locally or globally interact according to a radial basis potential function. The focus of this work is on nding the equilibrium conguration of the system, the ground state, by using Monte Carlo simulation method. The ground states can be of special interest in understanding the pattern formation in natural systems.
  Since the feasibility of the simulation could be challenged due to the high computational cost if the system size is large enough, the motivation for the current thesis is to reduce the above-mentioned computational complexity. For this aim, two eficient mathematical techniques, Fast Fourier Transform, and Fast Multipole Methods, are employed and O(N logN) complexity is achieved. The corresponding performance, features and feasibility of each method is investigated and discussed.

14.06.2012 - Stefan Kallenberger

• Title
  Compartmentalization and cell-to-cell variability in caspase-8 activation.
• Abstract
  Extrinsic apoptosis is initiated when a death inducing signaling complex (DISC) is formed at a plasma membrane death receptor upon ligand binding. Initiator caspases are activated by dimerization at the DISC, and then cleave themselves as well as proteins involved in mitochondria outer membrane permeabilization, and effector caspases, thus leading to cell death. Dynamical modeling is used to understand details of this molecular mechanism of apoptosis initiation and to get insights into heterogeneity in cleavage kinetics of caspases within a population of cells. To investigate the activation process of caspase-8, which is crucial to the decision to die or to survive as well as for the timing of apoptosis, model topologies reflecting different mechanisms of procaspase-8 activation were compared. To simultaneously record single cell and population observables under equal conditions, cell lines that stably express fluorescent cleavage probes for caspase-8 were generated from wild type and CD95 receptor overexpressing HeLa cells (CD95R-HeLa). By confocal live cell imaging and segmenting cellular compartments we generated single cell observables for membrane bound and cytosolic caspase-8. Western blot data representing mean protein concentrations inside a population of cells were measured for procaspase-8 and its fragments. For parameter identification and model discrimination, data were acquired from HeLa and CD95-HeLa cells at different cell death ligand concentrations. By bridging methods for single cell and population observables several boundary conditions were obtained that improved parameter estimations and discrimination between model topologies. This method facilitates the predictions of protein levels in single cells that can only be observed on the population level. Modeling with combined single cell and population data gave insight into cell-to-cell variability and signaling compartmentalization.

16.04.2012 - Johannes Bräuer

• Title
  An algorithm for exact likelihood calculation in dynamic Boolean networks with probabilistic time delays, and its application to structure learning
• Abstract
  Today the regulative function of many genes is still unknown. In order to elucidate causal relations, data from perturbation experiments is preferable over purely observational data. Researchers knock_out / knock_down different genes of a regulatory network to find out effects the genes resp. their gene products exer on each other. After such a perturbation, the expression level of these genes is measured at different time points. Even with this data it is still a major and largely unsolved challenge to infer the underlying regulatory mechanisms between the genes. Various approaches have been published, e.g. clustering of coregulated genes by using expression levels of genes at different time points of the experiment as a feature vector. In this work we present a novel, model-based approach using dynamic Boolean networks to address the problem of reconstructing the structure and the Boolean algebra, and the dynamic characteristics of small to medium-sized regulatory networks. Boolean networks are often appropriate for modeling regulatory dependencies because they seem to strike the right balance between the complexity of a model and its identifiability. Simulations show, that our approach can reconstruct the network structure of unknown networks with high accuracy. We apply our method to data of Ivanova et al. and a related network reconstruction attempt in Anchang et al. RNAi knockdown experiments were used to build a network of the key regulatory genes governing mouse stem cell maintenance resp. differentiation. Using our method can suggest some refinements (feedback loops) and corrections that lead to a better understanding of the dynamic interplay of some master regulators in embryonic stem cell development.

17.02.2012 - Benedikt Zacher

• Title
  Joint Bayesian Inference of miRNA and Transcription Factor Activities from Gene and microRNA Expression Data
• Abstract
  There have been many successful experimental and bioinformatics efforts to elucidate TF-target networks in several organisms. Attempts that use these networks in combination with gene expression data to draw conclusions on TF or miRNA activity are, however, still relatively sparse. BIRTA (Bayesian Inference of Regulation of Transcriptional Activity) uses a Bayesian network together with Markov Chain Monte Carlo (MCMC) sampling, to model and jointly infer TF and miRNA activities, from combined miRNA and mRNA expression data. Simulations reveal its good prediction performance in comparison to other approaches. Furthermore, the utility of BIRTA is demonstrated at the example of E. coli data comparing aerobic and anaerobic growth conditions, and by human expression data from healthy and cancerous pancreas tissue.

17.11.2011 - Melanie Boerris

Freiburg Institute for Advanced Studies (FRIAS), Center for Biological Systems Analysis, Freiburg, Germany

• Title
  Deciphering in vitro cell-cell communication between keratinocytes and firoblasts using communication theoretic approaches
• Abstract
  Tissue homeostasis and cell function in a multicellular organism are - to a great extend - defined by the interaction of cells with their environment, consisting of the extracellular matrix and surrounding cells. Cells communicate via soluble factors, called cytokines that act synergistically in space and time to which cells respond by changing their homeostasis and/or behavior. In general, cells communicate via double-paracrine cytokine feedback, in which all cells receive, process and secrete multiple cytokines that lead to a complex time-distributed stimulus-response action. While it is experimentally possible to access cytokine input factors into the cells and to monitor the whole cell response in terms of cell-wide changes in gene expression and de novo secretion of proteins, it is currently impossible to elucidate the relationship between complex cytokine stimulation, whole cell and cytokine response. This makes it hard to develop models predicting points of interference modulating cell communication. Here, we propose to experimentally decompose feedback-entangled cell-cell communication between normal human cytokines, and dermal fibroblasts using a mixed model approach. Statistical analysis of model error will help to elucidate the cause and effect and interaction of cytokines in this communication process. By iteratively increasing the complexity of cell communication from single stimulus, single cell to multiple input, multiple output in a double paracrine cell-cell communication setting, it will be the ultimate goal to unravel cellular signal processes that determine and control cell homeostasis and cell fate decisions, respectively.

16.11.2011 - Hauke Busch

Computational Biology, Center for Biological Systems Analysis, Freiburg, Germany

• Title
  Using Transcriptome Kinetics to Decipher Cellular Decisions - What do we see, What don’t we see?
• Abstract
  Cells initiate and control decisions like migration, proliferation or differentiation through an intricate, yet coordinated, regulation of large gene interaction networks. Here we show that network topology plays an important role in the cellular regulation, imposing constraints on gene regulation. Using in silico stimulus-response simulations of E. Coli and Yeast gene networks we find that highly connected network hubs genes are responding weakly, while strongly responding genes have, on average, a low degree of network connectivity. Being furthermore located at the network periphery, the latter act as effector genes, tightly linking to the cellular phenotype, being under the control of the moderately responding hub genes. As network topology is mostly conserved between species, a similar topology-dynamics relationship is expected in higher organisms. Hence, we applied our approach to migrating primary human keratinocytes under Hepatocyte Growth Factor stimulation as well as to dedifferentiating cells in the moss Physcomitrella Patens. Analysis of time-resolved microarray of migrating keratinocytes revealed a strong correlation between differentially regulated genes and cell migration. When inhibiting strongly responding genes, a decrease in the migratory activity proportional to the genes' response strength was found, in line with our initial hypothesis. To identify hub genes mediating dedifferentiation in P. Patens, we applied the idea of cell attractors in the search for genes that contribute to the coordinated, long-term change in gene expression, albeit responding moderately strong to the stimulus. The analysis highlighted nine novel transcription factors that possibly contribute to dedifferentiation, two of which have been experimentally verified already.

10.10.2011 - Jan Hasenauer

Institute for Systems Theory and Automatic Control, University of Stuttgart

• Title
  A unified framework for modeling and analysis of proliferating cell populations
• Abstract
  Background: Cell proliferation plays an essential role in most biological processes. Therefore, a multitude of mathematical models has been developed to describe proliferation processes, covering intracellular signal transduction as well as the population balance. In this work we focus on population models.
  
  Results: We present a unifying modeling and computational framework for proliferating cell populations. In contrast to existing models, the proposed model incorporates a discrete age structure as well as continuous label dynamics. Therefore, division number dependent parameters can be considered and the model prediction can also be directly compared to labeling experiments, e.g., with Carboxyfluorescein succinimidyl ester (CFSE). While the resulting coupled partial differential equations (PDEs) is highly complex, we prove that it can be decomposed into a system of ordinary differential equations (ODEs) and a set of decoupled PDEs. This reduces the computational effort tremendously. Furthermore, the PDEs are solved analytically and the ODE system is truncated, which allows simulation using a low-dimensional system of ODEs.
  
  Conclusion: The unified modeling approach in combination with the analytical and numerical treatment provides several advantages compared to existing approaches. These advantages are illustrated by studying the proliferation dynamics of immune cells.

1.8.2011 - Dr. Gunnar Cedersund

Linköping center of Systems Biology (LCSB), ISB group, Linköping University

• Title
  Evolving systems biology from suggestions to conclusions: methods and examples from type 2 diabetes
• Abstract
  Systems biology and its usage of mathematical modelling may well end up revolutionizing biology, since modelling can much better deal with complexity in data and underlying explanations. However, so far, most models remain under-determined by experimental data, and model predictions are therefore just suggestions, with a potentially arbitrarily high uncertainty. Here I will discuss methods - including e.g. identifiability analysis plus model reduction, set-based approaches, or modified optimization - that allows you to identify those predictions that must be fulfilled if the given model should describe the given data. These uniquely identified core predictions together with model rejections are stronger statements than suggestions; they are conclusions. I will use examples from real projects where such conclusive modelling has been used as an integrated part of data analysis to draw conclusions that could not have been drawn without modelling. These examples primarily regard insulin signalling and multi-level modelling of type 2 diabetes. (e.g., Brännmark et al, Nyman et al, JBC, 2010/2011). The talk is meant to be understandable yet interesting for both biologists and theoreticians. 

29.7.2011 - Felix Buggenthin

• Title
  Computational Prediction of Hematopoietic Cell Fates Using Single Cell Time Lapse Imaging
• Abstract
  Stem cells are able to regenerate and to give rise to specialized cell types. Due to these unique properties, stem cells represent a huge opportunity in the treatment of severe diseases such as dementia or leukemia. The hematopoietic stem cells residing in the bone marrow are the origin of blood regeneration and have been studied for many years. However, the processes driving hematopoiesis on cellular as well as molecular level are not fully understood. A major question to be elucidated is, at which generation after  intering differentiation a hematopoietic progenitor cell is committing to the erytroid or myeloid lineage and which cellular factors are involved in this decision. Continuous imaging of living hematopoietic stem cells and all their progeny in vitro allows detailed insights into differentiation. The vast amount of data and information generated in these experiments render it necessary to apply computational methods for quantification and analysis. In this thesis, we present a method capable of predicting a hematopoietic stem cell’s decision to commit to erytroid or myeloid lineage. Based on experiments from the Institute
  of Stem Cell Research at the Helmholtz Zentrum Munich, an automated image processing pipeline is established. The pipeline analyses 14 morphological properties of hematopoietic stem cells and all their progeny in brightfield images. In addition, fluorescence intensities of eYFP-labeled PU.1 molecules, a transcription factor that is thought to play a key role in lineage decision, are quantified. All features, intensities and annotations per cell are then utilized to train a random forest classifier that is able to predict cells commited to
  one of the lineages. Evaluation by ten-fold cross-validation results in a macro-averaged f1-measure of 0.83. Based on randomly drawn differentiation trees it is shown, that fate prediction is applicable for formerly unknown trees. The methods developed in this thesis can easily be adapted on other cell types, which could allow detailed analysis of stem cell behavior over their entire lifetime.

29.7.2011 - Florian Büttner

Joint Department of Physics, Institute of Cancer Research and Royal Marsden NHS Foundation Trust

• Title
  Reducing side-effects after radiotherapy treatments of cancer: novel approaches to model the dose-response relationship of organs-at-risk and implications on the treatment-planning process.
• Abstract
  The challenge of radiotherapy treatment of cancer is to maximise the radiation dose to the tumour while minimising radiation-induced side-effects. These side effects occur due to the inevitable deposition of dose in the healthy tissue surrounding the tumour. In order to choose the best treatment for a patient it is crucial to understand the dose-response of organs at risk of complication. Past approaches to model the dose-response have been based on volumetric information such as the mean dose to an organ only. In this work we investigate the role of information on the spatial distribution of dose and intra-organ variations in radio-sensitivity.
  Beneficial dose-patterns which reduce the risk of side-effects after prostate radiotehrapy due to the dose to the rectum are identified. Novel treatment-planning guidelines are established and a non-linear interpretable predictive model is generated and validated.
  Xerostomia (severe dry mouth) due to the dose to salivary glands is a common side-effect after radiotherapy for head and neck cancer.  Bayesian model selection based on a reversible jump MCMC algorithm is used to identify dose-response models allowing for regional variations of radiosensitivity. In conclusion, beneficial dose patterns could be identified for the rectum and the parotid glands. Tools which allow integration of these novel insights into clinical practice were presented.

13.7.2011 - Justin Feigelman

ETH Zürich

• Title
  Accelerated Multiscale Methods for Stochastic Simulation of Chemical Reaction Networks
• Abstract
  This thesis presents a review of modern multiscale stochastic simulation algorithms and their application to model systems of chemical reactions.  Novel methods are presented that combine leaping algorithms with multiscale techniques to achieve increased performance in systems that satisfy certain conditions. A discussion of the algorithms, conditions, and performance for the new methods is presented, along with proposals for future improvements.
  
  

7.7.2011 - Debarka Sengupta

Indian statistical institute - Machine Intelligence Unit

• Title
  MicroRNAs-Transcription factors induced network in human
• Abstract
  It has now become widely accepted that microRNAs participate actively
  along with transcription factors (TFs) to weave an inter-regulatory
  network that controls the gene expressions. Human genome wide realization
  of such a network gives insight into the passive regulations among
  microRNAs. It also helps explain the indirect interactions among microRNAs
  and target genes. The topological analysis of the microRNA-TF induced
  network reveals some of its intrinsic characteristics like small world
  property, presence of the giant component etc. Novel topological overlap
  analysis helps extract many clusters of miRNAs and TFs having significant,
  common disease associations.

21.6.2011 - Nikola Müller

Max Planck Institute of Biochemistry - Cellular Dynamics and Cell Patterning

• Title
  Similarities in Cytomes and Interactomes Point to Cellular Mechanismus
• Abstract
  Biological studies across all omics fields generate vast amounts of data. To understand these complex data, data mining techniques are indispensable. Approaches differ greatly when being either biologically- or computationally-driven.
  A cytome screen suggested a fundamental model for self-organization of biological membranes based on interactions between proteins and their associated lipids. We systematically investigated distribution and dynamics  of the yeast plasma membrane (PM) proteins with TIRF microscopy. Remarkably, all examined proteins were distributed non-homogeneously in numerous coexisting protein domains with all possible degrees of overlap. The extent of co-localization between integral PM proteins correlated with the sequence similarity between their transmembrane segments (TMS), and we could predictably relocate proteins by swapping their TMS. Correct domain association of PM proteins was essential for their biological function.  
  An interactome dataset of genetic interactions was analyzed for functionally enriched subgraphs:  We developed a parameter-free graph clustering algorithm based on the concept of graph compression in order to find sets of highly interlinked genes with strong similarity. The novel clustering algorithm for weighted graphs was based on the Minimum Description Length (MDL) principle in combination with a bisecting k-Means strategy. MDL relates the clustering problem to the problem of data compression: A good cluster structure on graphs enables strong graph compression, thus compression rates serve as similarity metrics for nodes. The novel clustering algorithm enriches group of genes in clusters sharing functional similarity.
  
  

12.5.2011 - Prof.Dr. Joachim Rädler

LMU Department für Physik - Soft condensed matter group

11.5.2011 - Dr. Julio Saez-Rodriguez

European Bioinformatics Institute (EBI)

14.4.2011 - Andreas Raue

University of Freiburg - Physics Institute

• Title
  Addressing Parameter Identifiability by Model-Based Experimentation
  
• Abstract
  Modeling of molecular reaction networks by ODEs faces difficulties when estimating model parameters from incomplete and noisy experimental data. By means of an application from cell biology (Becker, 2010) we present an approach that uses the profile likelihood to detect both structural and practical non-identifiabilities and investigates their influence on the model dynamics (Raue, 2009). The approach also allows to design new experiments that resolve non-identifiabilities and to derive confidence intervals.Becker V, Schilling M, Bachmann J, Baumann U, Raue A, Maiwald T, Timmer J and Klingmüller U (2010) Covering a broad dynamic range: information processing at the erythropoietin receptor.Science, 328(5984), 1404-1408Raue A, Kreutz C, Maiwald T, Bachmann J, Schilling M, Klingmüller U and Timmer J (2009) Structural and practical identifiability analysis of partially observed dynamical models by exploiting the profile likelihood. Bioinformatics, 25(15), 1923-1929

7.4.2011 - Dr. Saran Vardhsnabhuti

University of Pennsylvania

28.3.2011 - Dr. Nicolai Fricker

DKFZ Heidelberg

• Title
  The regulation of CD95-mediated cell death by c-FLIP - a Systems  Biology Approach
  
• Abstract
  C-FLIP proteins regulate caspase-8 activation and death receptor-induced apoptosis. The function of the short c-FLIP isoforms  (c-FLIPR and c-FLIPS) as inhibitors of cell death has been well  described. However the role of the long isoform of c-FLIP, c-FLIPL,  has been unclear. There are contradictory reports on whether c-FLIPL promotes or blocks cell death. Here, using a combination of mathematical modeling, imaging and quantitative western blots we elucidate the role of c-FLIPL in CD95-mediated cell death. Our data show that c-FLIPL can either act pro-apoptotic or anti-apoptotic, depending on the cellular context and the strength of receptor stimulation. Moreover we demonstrate that c-FLIPR and c-FLIPS act as a molecular switch deciding whether c-FLIPL blocks or promotes apoptosis. Our findings resolve the present controversial discussion on the function of c-FLIPL as a pro- and anti-apoptotic protein in death receptor-mediated death.