 |
||
 |
||
 |
||
| [Basel Computational Biology Conference 2005] |
| Abstracts |
|||
Keynote Lecture: Mass Spectrometry based Proteomics:
Computational Challenges and Partial Solutions
The objective of proteomics is the systematic analysis of the proteins expressed by a cell, tissue or organism. It is expected that such analyses will define comprehensive molecular signatures of tissues, cells and body fluids in health and disease. Such signatures are impacting a wide range of biological and clinical research questions, such as the systematic study of biological processes and the discovery of molecular clinical markers for detection, diagnosis and assessment of treatment outcome. The application of proteomics technology has proven particularly beneficial in cases in which differences between the proteomes (or fractions thereof) isolated from cells at different states have been analyzed, i.e. in which the analyses have been performed with accurate quantification. Currently most successful quantitative proteomic analyses are based on mass spectrometry and tandem mass spectrometry. In the context of such studies 10exp4 to 10exp5 tandem mass spectra are generated, each one potentially representing a unique peptide sequence. The computational assignment of these spectra to their corresponding peptide sequences, the statistical validation of these assignments, the extraction of reliable biological information from these datasets and the dissemination of the data represent a series of significant computational challenges that are at present only partially solved. In this presentation we will discuss current platforms for the mass spectrometric collection of proteomic data, describe a suite of OS source tools for their computational analysis and discuss remaining challenges Since most biological networks involve proteins, proteomics, the global analysis of the protein complement of a cell or tissue is a central element of systems biology. In this presentation we will discuss the current status of quantitative proteomics technologies and some of the resources that have emerged form the data they produce. We will also show with selected examples how quantitative proteomics can impact common types of experiments currently carried out in many biological research projects and discuss the challenges that remain to turn proteomics into a truly genomic science. References:
|
|||
|
Proteomics aims at deciphering the proteome, the complement of the genome, with the goal of increasing the understanding of biological processes, as well as improving and speeding up the development of drugs by discovering disease biomarkers and drug targets. The major elements of proteome analysis are powerful protein separation techniques such as liquid chromatography (LC) and two-dimensional electrophoresis (2-DE) associated to enzymatic processing and mass spectrometry (MS). These techniques have requested considerable efforts in the development of dedicated bioinformatics for over two decades, providing researchers with comprehensive and state-of-the-art software tools and databases for proteome analysis. The major areas of interest encompass protein identification from MS data and the storage and exchange of proteomics data. Current challenges include in particular the in-depth characterization of proteins from MS data using new and advanced algorithms, exploiting to its fullest the available experimental data, in particular data redundancy to extract biological knowledge, and the integration of information available across several databases as a step towards integrated systems biology.
|
|||
|
The wealth of high-quality functional genomics data sets and the improved accessibility of high-performance computing power have vastly propelled the development of new computational methods that are capable of integrating these large-scale heterogeneous data sets. Depending on the application purpose, one may select from a variety of methods with different analysis focus (e.g., clustering techniques, statistical replicate analysis, correlation analysis, neural nets). Additionally, genome-derived metabolic network models are now increasingly developed in particular for microbial systems, mostly due to their reduced biochemical complexity in comparison to higher-eukaryotic systems. These usually stoichiometric models are applied for studying the effects of genetic modification and change in environmental parameters, and the impact these modifications cause on metabolic flux distribution embedded into the cellular context. The ultimate vision of utilizing such models in the biotech industry is to gain a systems-level understanding of cellular physiology and thereby to assist in both rational strain engineering and process development strategies.
|
|||
As
cellular networks are getting more and more refined, it is becoming
feasible to move from 2D representations (nodes and edges) to 4D
i.e. explore temporal and spatial aspects of interaction networks.
I will introduce into recent work from our group to reveal temporal
changes (ranging from 90 minutes during the yeast cell cycle to
more than 2
|
|||
As current antibiotics therapy becomes increasingly ineffectual, new technologies are required to identify and develop novel classes of antibacterial agents. Our comparative genome analyses enabled prediction of novel cellular functions and complete pathways in bacteria, in order to characterise novel targets suitable for antibacterial compound screening. However, holistic strategies alternative to the focused target-based approach become more and more important in antibiotic drug discovery. Based on a compendium of genome-wide expression profiles reflecting the physiological response of the model bacterium Bacillus subtilis to hundred different antibiotic agents, we are able to simulate regulatory networks and pathways and to predict their genetic control elements. This way, we identified novel biomarkers for physiological stress states, suitable for screening of compounds with specific mechanisms of action. Moreover, our more elaborate expression profile analysis based on classification algorithms as well as regulon and pathway-specific data evaluation became a valuable tool to discover the mechanism of action of novel antibiotic agents.
|
|||
Wilhelm Gruissem [1], Anja Wille, Philip Zimmermann, Eva Vranova, Andreas Fürholz, Oliver Laule, Stefan Bleuler, Lars Hennig, Mattthias Hirsch-Hoffmann, Amela Prelic, Lothar Thiele, Eckart Zitzler and Peter Bühlmann, Reverse Engineering Group [2] and Functional Genomics Center Zurich [3], Swiss Federal Institute of Technology (ETH), Zurich. The analysis of genetic regulatory networks was greatly advanced by the availability of large data sets from high-throughput technologies such as DNA microarrays. The genome-wide, parallel monitoring of gene activity will increase our understanding of the molecular basis of pathway functions and their cellular network context. In simple eukaryotes or prokaryotes, gene expression data has been combined with two-hybrid data and phenotypic data to successfully predict protein-protein interaction and transcriptional regulation on a large scale. In higher organisms, however, little is known about regulatory control mechanisms and pathway networks on a larger scale. As a first step we have focused on isoprenoid metabolism, which is universally conserved and essential for cell survival. Arabidopsis has to independent pathways that function in the cytoplasm and chloroplast [4]. We developed a novel graphical Gaussian modelling (GGM) approach to elucidate the regulatory network of the two isoprenoid biosynthesis pathways bases on large scale expression data [5]. When applying this approach to infer a gene network, we detect modules of closely connected genes and candidate genes for cross-talk between the isoprenoid pathways. Genes of downstream pathways also fit well into the network. We evaluated our approach in a simulation study and using the yeast galactose utilization network. Connected genes were independently validated using Genevestigator [6], a novel powerful software suite for visualization of microarray and other data in their biological context. References:
|
|||
The accumulating wealth of genomes and other types of genomics data gives us the opportunity both to predict the function of proteins and their involvement in pathways as well as to trace the evolution of such biomolecular systems. As genomics data are however inherently noisy we need comparative analysis between multiple sets of data to make reliable predictions. We have shown that, while the co-expression of genes or the yeast-2-hybrid interaction of their proteins in one species only provides a weak signal that their proteins functionally interact, when that co-expression or interaction is measured in multiple species, it does become a reliable signal (van Noort et al, 2003; Huynen et al., 2004). One of the surprising observations of such “horizontal comparative genomics” between species, is the low level of conservation: less that 5% of genes that are co-expressed in are also co-expressed in C.elegans , and less than 25% of the yeast-2-hybrid interacting proteins from S.cerevisiae have been observed to interact in D.melanogaster . The question rises whether such low conservation reflects evolution and the changing relations between proteins or merely the noisy level of the datasets. When comparing yeast-2-hybrid data between species the level of conservation is only slightly lower than when comparing independently generated datasets from a single species, indicating that indeed the low reproducibility of genomics data might be the main cause for the low level of measured conservation between species. In order to filter out the noise from such analyses we have constructed a set of reliably co-regulated genes in S.cerevisiae by combining co-expression data with transcription factor binding data from ChIP-on-chip experiments. For those gene-pairs for which we have multiple sources of evidence that they are indeed truly co-regulated in S.cerevisiae , the conservation of co-regulation in C.elegans is 78%. Co-regulation therefore does appear well conserved in evolution (Snel et al., 2004). Such analyses however only apply to cases where both co-regulated genes are present in the species compared. Analyses of the phylogenetic distribution of proteins from a single biomolecular system indicate however suprisingly little “evolutionary modularity” of functional modules (Snel et al, 2004). By mapping such variation in the makeup of biomolecular systems on a phylogenetic tree one can actually reconstruct the evolution of biomolecular systems, a specific example of a large protein complex in eukaryotes will be discussed.
|
|||
Life sciences play
a crucial role in our society. Breakthroughs in fundamental research
are exploited at an increasingly higher rate and practically applied
in diagnostic, medical therapy and agro-food industry. It is imperative
that Switzerland avoids de-industrialization and re-organizes itself
in order to take on the challenges of the future and to make significant
efforts with the most important technologies and their applications
to life sciences, health care and nutrition.
|
|||
Proteomics is a key technology for the discovery of biomarkers that are required for pharmaceutical research and diagnostics. These markers can be found by massive parallel investigation of biological samples, preferably directly using diseased tissue. In order to obtain sufficient sensitivity, multidimensional protein fractionation schemes have to be employed, whereas statistical significance is achieved by the comparison of large numbers of samples. This strategy imposes limitations on the employed technologies. Thus, gel image comparison, as well as manual curation of mass spectrometric identification results are not feasible for large scale biomarker studies. We will show that meaningful data interpretation is possible only with high accuracy in protein identification so that false positive identifications will not obscure the true differences. In our group we have developed alternative solutions to the problem of protein quantification which employ data redundancies built into the experimental design of the biomarker study. Examples of successful biomarker discovery including the methodology for pre-validation and validation will be shown.
|
|||
In
my presentation Iwill discuss two applications of physical modeling
to biological systems. In the first, we model populations of cellular
oscillators to interpret recordings of a luciferase reporter in
a circadian cell culture assay. Correlation with single cell data
illustrates the complimentary of both techniques. Our analysis uncovered
reciprocal interactions between the circadian and cell cycle oscillators,
manifest for example as a gating of mitosis time by the clock.
|
|||
Brian Stoll, Anna Georgieva, Gabriel Helmlinger, Birgit Schoeberl, Tad Stewart, Ulrik Nielsen and Mark Penney The IGF network has been implicated in a number of cancers and therefore IGF1 and IGF1R have become promising targets for therapeutic intervention. In this work, a systems biology model of the IGF signalling pathway was produced in collaboration with Merrimack Pharmaceuticals Inc with the objective of identifying and quantifying biochemical biomarkers for the pharmacodynamic efficacy of an IGFR-1 inhibitor with clinical potential, and to further evaluate biomarkers for patient response. The IGF pathway topology was described using existing data in the literature and expressed as a mathematical model in Matlab. It was quantified by training it with measured in-house data in an iterative process which demonstrated the importance of having a high quality data set, in this case ones which described the peak activation of ERK and AKT well. This resulted in a model which predicted the downstream activation of ERK and AKT with a good degree of accuracy. (can you split this sentence into two) The model was then validated by comparing the predicted output in response to an IGF1-R inhibitor to an independently measured experimental set. Model
simulation and sensitivity analysis were used to determine those
biomarkers most sensitive to IGF1R blockade. These showed that the
IGF network lacks downstream signal amplification; consequently
the most sensitive biomarker is the phosphorylation of the IGF1R
itself, in contrast to similar pathways such as the EGF signalling
pathway. Model-based analysis also shows that normal expression
levels of IGF1R, levels of free IGF1 and IGF2, and IRS-1 expression
are the most important biomarkers for patient response. Furthermore,
it is shown that IGFBP-5 may also be important due to its ability
to amplify IGF signalling, illustrating that the regulation of free
IGF levels play a key role in IGF signalling prior to interaction
with IGF receptors.
|
|||
J.
Jeremy Rice, Lan Ma, John Wagner, and Gustavo Stolovitzky. IBM Computational
Biology Center, Yorktown Heights, NY. The
tumor suppressor p53 protein is critical to ensure genomic stability
when cells are under ionizing radiation (IR) stress. Recently it
was observed that single-cell response of p53 to IR is "digital",
in that it is number of oscillations (rather than the amplitude)
of p53 what shows dependence with the radiation dose. We present
a mathematical model of this phenomenon. In our model, double strand
break (DSB) sites induced by IR interact with a limiting pool of
DNA repair proteins, forming DSB-protein complexes at DNA damage
foci. Both the initial number of DSBs and the DNA repair process
are modeled taking into account the stochastic nature of the repair
process. The model assumes that the persisting complexes are sensed
by ataxia telangiectasia mutated (ATM), a kinase with a positive
feedback mechanism of autophosphorylation that sensitively transduces
the DNA damage information to downstream processes. The ATM sensing
module produces a step-like, ON-to-OFF signal as the input to a
downstream oscillator consisting of a p53-Mdm2 (Mdm2 is the negative
regulator of p53) autoregulatory feedback loop. Our simulation results
show that p53 and Mdm2 exhibit a coordinated oscillatory dynamics
upon IR stimulation, with a stochastic number of oscillations whose
mean increases with IR dosage, in good agreement with the observed
response of p53 to DNA-damage in single-cell experiments. We conjecture
that the robustnes of the oscillatory behavior of p53 is in part
induced by the ATM-induced autodegradation of MDM2, a mechanism
recently reported but not yet included in other models.
|
|||
Systems biology aims at understanding complex biological networks through a combination of (comprehensive) experimental analysis and (quantitative) mathematical modeling. At present, however, it is largely unclear, which knowledge and data will be required for establishing realistic mathematical models. Related to this, it is equally important to ask to what extent the already available data allow for meaningful model development. In this talk, I will argue that one can extract an unexpectedly high degree of information by appropriate combinations of modeling approaches, biological knowledge and only few experimental data. The examples presented will include (i) structural analysis of metabolic networks to infer key aspects of functionality and regulation, (ii) comparative computational modeling of the TOR (‘target of rapamycin') pathway to reveal signaling mechanisms, and (iii) detailed modeling of a complex network in yeast cell cycle regulation. These studies point to the robustness of cellular networks as an ‘enabling' feature for model development, and they suggest strategies for efficiently linking future experimental and theoretical approaches to cellular networks.
|
|||
In this presentation we give an overview of some computational challenges in the integrative analysis of transcriptomics, proteomics and metabolomics data in order to obtain a better understanding of cellular processes, a prerequisite for systems biology. We start by discussing some of the strengths and weaknesses pertaining to the different 'omics' levels, calling for the need to be able to work at more than one level. We continue with a few examples of how we at Genedata tackle some of those challenges. For example, how metabolic pathways can be studied in an integrated fashion together with expression analysis data of different types. We then move on to some issues that we find especially important to address in order to be able to make maximum use of the data, particularly the necessity for quality assessment of the raw data as well as the subsequent statistical analysis and data mining to extract the biological information. Finally, we illustrate the methods by applying them to the problem of biomarker identification.
|
|||
A proper understanding of the mechanisms controlling gene expression requires the integration of molecular and genetic data into full fledge mathematical models. An overview of the main dynamical modelling approach will be provided, before focusing on a multi-level, logical approach, which enables a flexible qualitative modelling of complex regulatory networks. This approach encompasses the development of a dedicated software suite (GIN-sim), and will be illustrated by applications to pattern formation and cell differentiation in the fly Drosophila melanogaster .
|
|||
The
face of biological research in the biotech industry has evolved
at an alarming rate. From a one-gene/one protein analysis it has
borne witness to a multitude of new technologies that allow us to
capture and integrate a vast amount of information generated by
high throughput methods such as DNA microarrays, proteomics and
bioinformatic. Then along has come the sequenced human genome, and
suddenly we have a complete skeleton upon which to integrate the
mass of information generated. The scientific community now
has an integrated way of looking at what have previously been isolated
snippets of knowledge. We have known for some time the function(s)
of many proteins in signaling pathways, developmental regulation,
cell cycle progression, and so on. However, what is becoming clearer
as we gather more information and gaze upon the global picture,
is that a single protein rarely performs a single function.
Rather, the activity that we assign to it is the product of its
interaction with other proteins, small molecules or nucleic acids
at any given time. Despite the advance in high throughput technologies
(or, perhaps, because of this), we are faced with an avalanche of
data but only flakes of knowledge. What is needed is a system approach
that would enable us to integrate all the information generated
from these technology platforms and develop both mathematical and
biological methodologies to test them.
|
|||
MicroRNAs
are a large class of endogenous RNA molecules approximately 22 nucleotides
in length that regulate translation of protein-coding genes in plants
and animals. Hundreds of miRNA genes have been identified in various
eukaryotic organisms and their number is still growing. Their existence
and function in “simpler” life forms such as viruses
have not been extensively investigated, although viruses are known
to use many elaborate RNA processing functions of the host. Using
a combination of computational miRNA gene prediction and small RNA
cloning, we discovered miRNAs encoded by herpes viruses. The predicted
viral targets and the expression profile of viral miRNAs suggest
a role of these miRNAs in the viral life cycle, and thus interaction
with the host.
|
|||
