Yes / This chapter presents the Infobiotics Workbench (IBW), an integrated software suite developed for computational systems biology. The tool is built upon stochastic P systems, a probabilistic extension of P systems, as modelling framework. The platform utilises computer-aided modelling and analysis of biological systems through simulation, verification and optimisation. IBW allows modelling and analysing not only cell level behaviour, but also multi-compartmental population dynamics. This enables comparing be tween macroscopic and mesoscopic interpretations of molecular interaction networks and investigating temporo-spatial phenomena in multicellular systems. These capabilities make IBW a useful, coherent and comprehensive in silico tool for systems biology research.

Thesis (MSc)--Stellenbosch University, 2011. / ENGLISH ABSTRACT: Progress in computational systems biology depends crucially on the availability of generic rate equations that accurately describe the behaviour and regulation of catalysed processes over a wide range of conditions. Such equations for ordinary enzyme-catalysed reactions have been developed in our group and have proved extremely useful in modelling metabolic networks. However, these networks link to growth and reproduction processes through template-directed synthesis of macromolecules such as polynucleotides and polypeptides. Lack of an equation that captures such a relationship led us to derive a generic rate equation that describes catalysed, template-directed polymerisation reactions with varying monomer stoichiometry and varying chain length. A model describing the mechanism of a generic template-directed polymerisation process in terms of elementary reactions with mass action kinetics was developed. Maxima, a computational algebraic solver, was used to determine analytical expressions for the steady-state concentrations of the species in the equation system from which a steady-state rate equation could be derived. Using PySCeS, a numerical simulation platform developed in our group, we calculated the time-dependent evolution and the steadystates of the species in the catalytic mechanisms used in the derivation of the rate equations. The rate equation was robust in terms of being accurately derived, and in comparison with the rates determined with PySCeS. Addition of more elongation steps to the mechanism allowed the generalisation of the rate equation to an arbitrary number of elongations steps and an arbitrary number of monomer types. To test the regulatory design of the system we incorporated the generic rate equation in a computational model describing a metabolic system consisting of multiple monomer supplies linked by a template-directed demand reaction. Rate characteristics were chosen to demonstrate the utility of the simplified generic rate equation. The rate characteristics provided a visual representation of the control and regulation profile of the system and showed how this profile changes under varying conditions. / AFRIKAANSE OPSOMMING: Die beskikbaarheid van generiese snelheidsvergelykings wat die gedrag en regulering van gekataliseerde prosesse akkuraat oor ’n wye reeks omstandighede beskryf is van kardinale belang vir vooruitgang in rekenaarmatige sisteembiologie. Sulke vergelykings is in ons groep ontwikkel vir gewone ensiem-gekataliseerde reaksies en blyk uiters nuttig te wees vir die modellering van metaboliese netwerke. Hierdie netwerke skakel egter deur templaat-gerigte sintese van makromolekule soos polinukleotiede en polipeptiede aan groei- en voorplantingsprosesse. Die gebrek aan vergelykings wat sulke verwantskappe beskryf het ons genoop om ’n generiese snelheidsvergelyking af te lei wat gekataliseerde, templaatgerigte polimerisasie-reaksies met wisselende monomeerstoigiometrie en kettinglengte beskryf. ’n Model wat die meganisme van ’n generiese templaat-gerigte polimerisasie-proses in terme van elementêre reaksies met massa-aksiekinetika beskryf is ontwikkel. Maxima, ’n rekenaarmatige algebraïese oplosser, is gebruik om analitiese uitdrukkings vir die bestendige- toestand konsentrasies van die spesies in die vergelyking-stelsel te vind. Hierdie uitdrukkings is gebruik om ’n bestendige-toestand snelheidsvergelyking af te lei. Ons het die tyd-afhanklike progressie en die bestendige toestande bereken van die spesies in die katalitiese meganismes wat gebruik is in die afleiding van die snelheidsvergelykings. Die rekenaarprogram PySCeS is ’n numeriese simulasieplatform wat in ons groep ontwikkel is. Die snelheidsvergelyking blyk akkuraat afgelei te wees en is in ooreenstemming met snelhede deur PySCeS bereken. Die toevoeging van verdere verlengingstappe tot die meganisme het dit moontlik gemaak om die snelheidsvergelyking te veralgemeen tot ’n arbitrêre hoeveelheid verlengingstappe en monomeertipes. Om die regulatoriese ontwerp van die sisteem te toets het ons die generiese snelheidsvergelyking in ’n rekenaarmatige model geïnkorporeer wat ’n metaboliese sisteem bestaande uit verskeie monomeer-aanbodblokke en ’n templaatgerigte aanvraagblok beskryf. Snelheidskenmerkanalise is gekies om die nut van die vereenvoudigde generiese snelheidsvergelyking te demonstreer. Met hierdie snelheidskenmerke kon ons die kontrole- en reguleringsprofiel van die stelsel visualiseer en wys hoe hierdie profiel verander onder wisselende omstandighede.

Computational systems biology is an emerging area of research that focuses on understanding the holistic view of complex biological systems with the help of statistical, mathematical and computational techniques. The regulation of gene expression in gene regulatory network is a fundamental task performed by all known forms of life. In this subsystem, modelling the behaviour of the components and their interactions can provide useful biological insights. Statistical approaches for understanding biological phenomena such as gene regulation are proving to be useful for understanding the biological processes that are otherwise not comprehensible due to multitude of information and experimental difficulties. A combination of both the experimental and computational biology can potentially lead to system level understanding of biological systems. This thesis focuses on the problem of inferring the dynamics of gene regulation from the observed output of gene expression. Understanding of the dynamics of regulatory proteins in regulating the gene expression is a fundamental task in elucidating the hidden regulatory mechanisms. For this task, an initial fixed structure of the network is obtained using experimental biology techniques. Given this network structure, the proposed inference algorithms make use of the expression data to predict the latent dynamics of transcription factor proteins. The thesis starts with an introductory chapter that familiarises the reader with the physical entities in biological systems; then we present the basic framework for inference in transcriptional regulation and highlight the main features of our approach. Then we introduce the methods and techniques that we use for inference in biological networks in chapter 2; it sets the foundation for the remaining chapters of the thesis. Chapter 3 describes four well-known methods for inference in transcriptional regulation with pros and cons of each method. Main contributions of the thesis are presented in the following three chapters. Chapter 4 describes a model for inference in transcriptional regulation using state space models. We extend this method to cope with the expression data obtained from multiple independent experiments where time dynamics are not present. We believe that the time has arrived to package methods like these into customised software packages tailored for biologists for analysing the expression data. So, we developed an open-sources, platform independent implementation of this method (TFInfer) that can process expression measurements with biological replicates to predict the activities of proteins and their influence on gene expression in gene regulatory network. The proteins in the regulatory network are known to interact with one another in regulating the expression of their downstream target genes. To take this into account, we propose a novel method to infer combinatorial effect of the proteins on gene expression using a variant of factorial hidden Markov model. We describe the inference mechanism in combinatorial factorial hidden model (cFHMM) using an efficient variational Bayesian expectation maximisation algorithm. We study the performance of the proposed model using simulated data analysis and identify its limitation in different noise conditions; then we use three real expression datasets to find the extent of combinatorial transcriptional regulation present in these datasets. This constitutes chapter 5 of the thesis. In chapter 6, we focus on problem of inferring the groups of proteins that are under the influence of same external signals and thus have similar effects on their downstream targets. Main objectives for this work are two fold: firstly, identifying the clusters of proteins with similar dynamics indicate their role is specific biological mechanisms and therefore potentially useful for novel biological insights; secondly, clustering naturally leads to better estimation of the transition rates of activity profiles of the regulatory proteins. The method we propose uses Dirichlet process mixtures to cluster the latent activity profiles of regulatory proteins that are modelled as latent Markov chain of a factorial hidden Markov model; we refer to this method as DPM-FHMM. We extensively test our methods using simulated and real datasets and show that our model shows better results for inference in transcriptional regulation compared to a standard factorial hidden Markov model. In the last chapter, we present conclusions about the work presented in this thesis and propose future directions for extending this work.

Technological advances in the recent decades have enabled cancer researchers to probe the disease at multiple resolutions. This wealth of experimental data combined with computational systems biology methods is now leading to predictive models of cancer progression and response to therapy. We begin by presenting our research group’s multis-cale in silico framework for modeling cancer, whose core is a tissue-scale computational model capable of tracking the progression of tumors from a diffusion-limited avascular phase through angiogenesis, and into invasive lesions with realistic, complex morphologies. We adapt this core model to consider the delivery of systemically-administered anticancer agents and their effect on lesions once they reach their intended nuclear target. We calibrate the model parameters using in vitro data from the literature, and demonstrate through simulation that transport limitations affecting drug and oxygen distributions play a significant role in hampering the efficacy of chemotherapy; a result that has since been validated by in vitro experimentation. While this study demonstrates the capability of our adapted core model to predict distributions (e.g., cell density, pressure, oxygen, nutrient, drug) within lesions and consequent tumor morphology, nevertheless, the underlying factors driving tumor-scale behavior occur at finer scales. What is needed in our multi-scale approach is to parallel reality, where molecular signaling models predict cellular behavior, and ultimately drive what is seen at the tumor level. Models of signaling pathways linked to cell models are already beginning to surface in the literature. We next transition our research to the molecular level, where we employ data mining and bioinformatics methods to infer signaling relationships underlying a subset of breast cancer that might benefit from targeted therapy of Androgen Receptor and associated pathways. Defining the architecture of signaling pathways is a critical first step towards development of pathways models underlying tumor models, while also providing valuable insight for drug discovery. Finally, we develop an agent-based, cell-scale model focused on predicting motility in response to chemical signals in the microenvironment, generally accepted to be a necessary feature of cancer invasion and metastasis. This research demonstrates the use of signaling models to predict emergent cell behavior, such as motility. The research studies presented in this dissertation are critical steps towards developing a predictive, in silico computational model for cancer progression and response to therapy. Our Laboratory for Computational & Predictive Oncology, in collaboration with research groups throughout in the United States and Europe are following a computational systems biology paradigm where model development is fueled by biological knowledge, and model predictions are refining experimental focus. The ultimate objective is a virtual cancer simulator capable of accurately simulating cancer progression and response to therapy on a patient-specific basis. / text

By increasing the level of abstraction in the representation of regulatory network models, we can hope to allow modelers to create models that are beyond the threshold of what can currently be expressed reliably. As hundreds of reactions are difficult to understand, maintain, and extend, thousands of reactions become next to impossible without any automation or aid. Using the multistate-species concept we can reduce the number of reactions needed to represent certain systems and thus, lessen the cognitive load on modelers. A multistate species is an entity with a defined range for state variables, which refers to a group of different forms for a specific species. A multistate reaction involves one or more multistate species and compactly represents a group of similar single reactions. In this work, we have extended JCMB (the JigCell Model Builder) to comply with multistate species and reactions modeling and presented a proposal for enhancing SBML (the Systems Biology Markup Language) standards to support multistate models. / Master of Science

Hearkening to calls from life scientists for aid in interpreting rapidly-growing repositories of data, the fields of bioinformatics and computational systems biology continue to bear increasingly sophisticated methods capable of summarizing and distilling pertinent phenomena captured by high-throughput experiments. Techniques in analysis of genome-wide gene expression (e.g., microarray) data, for example, have moved beyond simply detecting individual genes perturbed in treatment-control experiments to reporting the collective perturbation of biologically-related collections of genes, or "processes". Recent expression analysis methods have focused on improving comprehensibility of results by reporting concise, non-redundant sets of processes by leveraging statistical modeling techniques such as Bayesian networks. Simultaneously, integrating gene expression measurements with gene interaction networks has led to computation of response networks--subgraphs of interaction networks in which genes exhibit strong collective perturbation or co-expression. Methods that integrate process annotations of genes with interaction networks identify high-level connections between biological processes, themselves. To identify context-specific changes in these inter-process connections, however, techniques beyond process-based expression analysis, which reports only perturbed processes and not their relationships, response networks, composed of interactions between genes rather than processes, and existing techniques in process connection detection, which do not incorporate specific biological context, proved necessary. We present two novel methods which take inspiration from the latest techniques in process-based gene expression analysis, computation of response networks, and computation of inter-process connections. We motivate the need for detecting inter-process connections by identifying a collection of processes exhibiting significant differences in collective expression in two liver tissue culture systems widely used in toxicological and pharmaceutical assays. Next, we identify perturbed connections between these processes via a novel method that integrates gene expression, interaction, and annotation data. Finally, we present another novel method that computes non-redundant sets of perturbed inter-process connections, and apply it to several additional liver-related data sets. These applications demonstrate the ability of our methods to capture and report biologically relevant high-level trends. / Ph. D.

The ever-growing size and complexity of molecular network models makes them difficult to construct and understand. Modifying a model that consists of tens of reactions is no easy task. Attempting the same on a model containing hundreds of reactions can seem nearly impossible. We present the JigCell Model Connector, a software tool that supports large-scale molecular network modeling. Our approach to developing large models is to combine together smaller models, making the result easier to comprehend. At the base, the smaller models (called modules) are defined by small collections of reactions. Modules connect together to form larger modules through clearly defined interfaces, called ports. In this work, we enhance the port concept by defining different types of ports. Not all modules connect together the same way, therefore multiple connection options need to exist. / Master of Science / Genes and proteins interact to control the functions of a living cell. In order to better understand these interactions, mathematical models can be created. A model is a representation of a cellular function that can be simulated on a computer. Results from the simulations can be used to gather insight and drive the direction of new laboratory experiments. As new discoveries are made, mathematical models continue to grow in size and complexity. We present the JigCell Model Connector, a software tool that supports large-scale molecular network modeling. Our approach to developing large models is to combine together smaller models, making the result easier to comprehend. At the base, the smaller models (called modules) are defined by small collections of reactions. Modules connect together to form larger modules through clearly defined interfaces, called ports. In this work, we enhance the port concept by defining different types of ports. Not all modules connect together the same way, therefore multiple connection options need to exist.

Background Molecular data of histologically classified oligodendrogliomas are available offering the possibility to stratify these human brain tumors into clinically relevant molecular subtypes. Methods Gene copy number, mutation, and expression data of 193 histologically classified oligodendrogliomas from The Cancer Genome Atlas (TCGA) were analyzed by well-established computational approaches (unsupervised clustering, statistical testing, network inference). Results We applied hierarchical clustering to tumor gene copy number profiles and revealed three molecular subgroups within histologically classified oligodendrogliomas. We further screened these subgroups for molecular glioma markers (1p/19q co-deletion, IDH mutation, gain of chromosome 7 and loss of chromosome 10) and found that our subgroups largely resemble known molecular glioma subtypes. We excluded glioblastoma-like tumors (7a10d subgroup) and derived a gene expression signature distinguishing histologically classified oligodendrogliomas with concurrent 1p/19q co-deletion and IDH mutation (1p/19q subgroup) from those with predominant IDH mutation alone (IDHme subgroup). Interestingly, many signature genes were part of signaling pathways involved in the regulation of cell proliferation, differentiation, migration, and cell-cell contacts. We further learned a gene regulatory network associated with the gene expression signature revealing novel putative major regulators with functions in cytoskeleton remodeling (e.g. APBB1IP, VAV1, ARPC1B), apoptosis (CCNL2, CREB3L1), and neural development (e.g. MYTIL, SCRT1, MEF2C) potentially contributing to the manifestation of differences between both subgroups. Moreover, we revealed characteristic expression differences of several HOX and SOX transcription factors suggesting the activity of different glioma stemness programs in both subgroups. Conclusions We show that gene copy number profiles alone are sufficient to derive molecular subgroups of histologically classified oligodendrogliomas that are well-embedded into general glioma classification schemes. Moreover, our revealed novel putative major regulators and characteristic stemness signatures indicate that different developmental programs might be active in these subgroups, providing a basis for future studies.

Background Molecular data of histologically classified oligodendrogliomas are available offering the possibility to stratify these human brain tumors into clinically relevant molecular subtypes. Methods Gene copy number, mutation, and expression data of 193 histologically classified oligodendrogliomas from The Cancer Genome Atlas (TCGA) were analyzed by well-established computational approaches (unsupervised clustering, statistical testing, network inference). Results We applied hierarchical clustering to tumor gene copy number profiles and revealed three molecular subgroups within histologically classified oligodendrogliomas. We further screened these subgroups for molecular glioma markers (1p/19q co-deletion, IDH mutation, gain of chromosome 7 and loss of chromosome 10) and found that our subgroups largely resemble known molecular glioma subtypes. We excluded glioblastoma-like tumors (7a10d subgroup) and derived a gene expression signature distinguishing histologically classified oligodendrogliomas with concurrent 1p/19q co-deletion and IDH mutation (1p/19q subgroup) from those with predominant IDH mutation alone (IDHme subgroup). Interestingly, many signature genes were part of signaling pathways involved in the regulation of cell proliferation, differentiation, migration, and cell-cell contacts. We further learned a gene regulatory network associated with the gene expression signature revealing novel putative major regulators with functions in cytoskeleton remodeling (e.g. APBB1IP, VAV1, ARPC1B), apoptosis (CCNL2, CREB3L1), and neural development (e.g. MYTIL, SCRT1, MEF2C) potentially contributing to the manifestation of differences between both subgroups. Moreover, we revealed characteristic expression differences of several HOX and SOX transcription factors suggesting the activity of different glioma stemness programs in both subgroups. Conclusions We show that gene copy number profiles alone are sufficient to derive molecular subgroups of histologically classified oligodendrogliomas that are well-embedded into general glioma classification schemes. Moreover, our revealed novel putative major regulators and characteristic stemness signatures indicate that different developmental programs might be active in these subgroups, providing a basis for future studies.

This thesis encompasses research on Artificial Intelligence in support of automating scientific discovery in the fields of biology and medicine. At the core of this research is the ongoing development of a general-purpose artificial intelligence framework emulating various facets of human-level intelligence necessary for building cross-domain knowledge that may lead to new insights and discoveries. To learn and build models in a data-driven manner, we develop a general-purpose learning framework called Syntactic Nonparametric Analysis of Complex Systems (SYNACX), which uses tools from Bayesian nonparametric inference to learn the statistical and syntactic properties of biological phenomena from sequence data. We show that the models learned by SYNACX offer performance comparable to that of standard neural network architectures. For complex biological systems or processes consisting of several heterogeneous components with spatio-temporal interdependencies across multiple scales, learning frameworks like SYNACX can become unwieldy due to the the resultant combinatorial complexity. Thus we also investigate ways to robustly reduce data dimensionality by introducing a new data abstraction. In particular, we extend traditional string and graph grammars in a new modeling formalism which we call Simplicial Grammar. This formalism integrates the topological properties of the simplicial complex with the expressive power of stochastic grammars in a computation abstraction with which we can decompose complex system behavior, into a finite set of modular grammar rules which parsimoniously describe the spatial/temporal structure and dynamics of patterns inferred from sequence data.