Challenges in scientific research include the difficulty in obtaining overviews of the large amount of data required for analysis, and in resolving the differences in terminology used to store and interpret information in multiple, independently created data sets. Ontologies provide one solution for analysis involving multiple data sources, improving cross-referencing and data integration. This thesis looks at harnessing advanced human perception to reduce the cognitive load in the analysis of the multiple, complex data sets the bioinformatics user group studied use in research, taking advantage also of users’ domain knowledge, to build mental models of data that map to its underlying structure. Guided by a user-centred approach, prototypes were developed to provide a visual method for exploring users’ information requirements and to identify solutions for these requirements. 2D and 3D node-link graphs were built to visualise the hierarchically structured ontology data, to improve analysis of individual and comparison of multiple data sets, by providing overviews of the data, followed by techniques for detailed analysis of regions of interest. Iterative, heuristic and structured user evaluations were used to assess and refine the options developed for the presentation and analysis of the ontology data. The evaluation results confirmed the advantages that visualisation provides over text-based analysis, and also highlighted the advantages of each of 2D and 3D for visual data analysis.

No description available.

Path integral formalism is a powerful tool borrowed from theoretical physics to build dynamical descriptions, yet its potential is largely unexplored in the context of complex networks, such as the ones common in systems biology. In this PhD thesis, I present di erent mathematical frameworks based on path integrals to capture the time evolution of interacting continuous degrees of freedom, e.g. biochemical concentrations. The generality of path integral approaches enables us to tackle several questions related to modelling and inference for dynamics. We first develop a novel mean field approximation, the Extended Plefka Expansion, for stochastic di erential equations exhibiting generic nonlinearities. The key element is the definition of “e ective” fields which map an interacting dynamics into the “most similar” non-interacting picture, i.e. the one producing the same average observables. In the resulting picture, couplings between variables are replaced by a memory and a coloured noise. We next apply this setup to the case in which part of the network is observed and part is unknown. We study the accuracy of prediction of the unobserved dynamics as a function of the number of observed nodes and other structural parameters of the system. The Extended Plefka Expansion is expected to become exact in the limit of infinite size networks with couplings of mean field type, i.e. weak and long-ranged. We show this explicitly for a linear dynamics by comparison with other methods relying on Random Matrix Theory. We finally appeal to path integrals to design “reduced” models, where equations are referred solely to some selected variables (subnetwork) but still carry information on the whole network. This model reduction strategy leads to substantially higher quantitative accuracy in the prediction of subnetwork dynamics, as we demonstrate with an example from the protein interaction network around the Epidermal Growth Factor Receptor.

No description available.

No description available.

Our life is being increasingly quantified by data. To obtain information from quantitative data, we need to develop various analysis methods, which can be drawn from diverse fields, such as computer science, information theory and statistics. This thesis focuses on investigating methods for analysing data generated for medical research. Its focus is on the purpose of using various data to quantify patients for personalized treatment. From the perspective of data type, this thesis proposes analysis methods for the data from the fields of Bioinformatics and medical imaging. We will discuss the need of using data from molecular level to pathway level and also incorporating medical imaging data. Different preprocessing methods should be developed for different data types, while some post-processing steps for various data types, such as classification and network analysis, can be done by a generalized approach. From the perspective of research questions, this thesis studies methods for answering five typical questions from simple to complex. These questions are detecting associations, identifying groups, constructing classifiers, deriving connectivity and building dynamic models. Each research question is studied in a specific field. For example, detecting associations is investigated for fMRI signals. However, the proposed methods can be naturally extended to solve questions in other fields. This thesis has successfully demonstrated that applying a method traditionally used in one field to a new field can bring lots of new insights. Five main research contributions for different research questions have been made in this thesis. First, to detect active brain regions associated to tasks using fMRI signals, a new significance index, CR-value, has been proposed. It is originated from the idea of using sparse modelling in gene association study. Secondly, in quantitative Proteomics analysis, a clustering based method has been developed to extract more information from large scale datasets than traditional methods. Clustering methods, which are usually used in finding subgroups of samples or features, are used to match similar identities across samples. Thirdly, a pipeline originally proposed in the field of Bioinformatics has been adapted to multivariate analysis of fMRI signals. Fourthly, the concept of elastic computing in computer science has been used to develop a new method for generating functional connectivity from fMRI data. Finally, sparse signal recovery methods from the domain of signal processing are suggested to solve the underdetermined problem of network model inference.

As the number of fully sequenced genomes grows, efforts are shifted towards investigation of functional aspects. One research focus is the transcriptome, the set of all transcribed genomic features. We aspire to understand what features constitute the transcriptome, in which context these are transcribed and how their transcription is regulated. Studies that aim to answer these questions frequently make use of high-throughput technologies that allow for investigation of multiple genomic regions, or transcribed copies of genomic regions, in parallel. In this dissertation, I present three high-throughput studies I have been involved in, in which data gained from oligo-nucleotide tiling microarrays or large-scale cDNA sequencing provided insights into the transcriptome and transcriptional regulation in the model organisms Saccharomyces cerevisiae and Mus musculus. Interpretation of such high-throughput data poses two major computational tasks. The primary statistical analysis includes quality assessment, data normalisation and identification of significantly affected targets, i.e. regions of the genome deemed transcribed or involved in transcriptional regulation. Second, in an integrative bioinformatic analysis, the identified targets need to be interpreted in context of the current genome annotation and related experimental results. I provide details of these individual steps as they were conducted in the three studies. For both primary and integrative analysis, functional, extensible and welldocumented software is required, which implements individual analysis steps, allows for concise visualisation of intermittent and final results and facilitates the construction of automated, programmed workflows. Ideally such software is optimised with respect to scalability, reproducibility and methodical scope of the analyses. This dissertation contains details of two such software packages in the Bioconductor project, which I (co-)developed.

Recent advances in sequencing technologies have brought a renewed impetus to the development of bioinformatics tools necessary for sequence processing and analysis. Along with the constant requirement to be able to assemble more complex genomes from ever evolving sequencing experiments and technologies there also exists a lack in visually accessible representations of information generated by analysis tools. Most of the novel algorithms, specifically for de novo genome assembly of next generation sequencing (NGS) data, are not able to efficiently handle data generated on large populations. We have assessed the common methods for genome assembly used today both from a theoretical point of view and their practical implementations. In this dissertation we present StarK (stands for k�), a novel assembly algorithm with a new data structure designed to overcome some of the limitations that we observed in established methods enabling higher quality NGS data processing. The StarK approach structurally combines de Brujin graphs for all possible dimensions in one supergraph. Although the technique to join reads remains in concept the same, the dimension k is no longer fixed. StarK is designed in such a way that it allows the assembler to dynamically adjust the de Brujin graph dimension k on the fly and at any given nucleotide position without losing connections between graph vertices or doing complicated calculations. The new graph uses localised coverage difference evaluation to create connected sub graphs which allows higher resolution of genomic differences and helps differentiate errors from potential variants within the sequencing sample. In addition to this we present a bioinformatics analysis pipeline for high-variation viral population analysis (including transmission studies), which, using both new and established methods, creates easily interpretable visual representations of the underlying data analysis. Together we provide a solid framework for biologists for extracting more information from sequencing data with less effort and faster than before.

No description available.

No description available.