Integrative Analysis & Modeling

Program Committee

Danielle Posthuma (program leader)
Mathisca de Gunst (program leader)



Niels Cornelisse
Jaap Heringa
Klaus Linkenkaer-Hansen
Pim van Nierop

Arjen van Ooyen
Sophie van der Sluis
Aad van der Vaart

Rationale

The general mission of the Neuroscience Campus is to study the brain and its disease mechanisms through an integrative systems biology approach running from molecule-to-bedside. Such an integrative systems biology approach includes experiments that range from cellular and molecular studies to whole brain imaging and whole genome studies. These studies generate large quantities of data, which are continuously increasing at higher levels of resolution. The data produced are heterogeneous, as they derive from different disciplines and different levels of study. As a key element in understanding the nervous system lies in integrating data derived from different disciplines and study levels, methodological approaches are needed that facilitate this integration. The core aim of this program is to advance the integration of neuroscience with information technology by developing mathematical and computational models and analysis tools for the sharing, integration and analysis of experimental neuroscience data, discovery research and the advancement of theories of nervous system function.

Background

Central to the integrative systems biology approach is the modeling cycle: the iterative cycle from experiment, data integration, modeling, and prediction. For different neuroscience research questions this cycle is passed through in different ways. In general, two systems biology approaches can be distinguished, top-down and bottom-up. On the one hand, large data sets are being produced to gain insight into the relationships between the different data types measured on different levels or scales. Statistics and bioinformatics are essential for the analysis of the large data sets produced by this top-down approach. On the other hand, model systems with a limited number of components are studied to unravel generic principles of the involved biological networks. For this bottom-up approach, biology-based mathematical and computational modeling, and model-based statistical data analysis can make important contributions to the modeling and prediction steps of the modeling cycle. Common to both approaches is the need for innovative integrative methods that combine techniques from different areas of mathematics and bioinformatics. Naturally, all method development will be performed in frequent consultation with researchers from the experimental groups of the Neuroscience Campus. These characteristics of the systems biology approach and the role of modeling and analysis in it, are reflected in the program themes.

Methods for Large Data Sets

Crossing disciplinary boundaries necessitates the development of novel research methods for large data sets. In particular, questions of a deeper, system-biological nature (e.g. time-dependence, spatial variation, networks and pathways, interactions) require integrated modeling and statistical techniques and a combination of these with bioinformatics algorithms and neuroinformatics databases. Combining statistics and bioinformatics approaches is therefore a key focus within this theme. For example, to determine significance of bioinformatics results, statistical expertise is needed. Ongoing research into the role of synaptic gene networks in complex traits and diseases has produced a semi-automated algorithm to integrate in silico gene information, animal genetic studies and human genome data. Other ongoing integrative studies are the research on gene-regulation of neuronal outgrowth, on the analysis of combined EEG/fMRI signals, and on the development of methods for estimation and testing  linkage disequilibrium from sib data. Often novel uses of combinations of statistical and computational techniques such as MCMC, Bayesian statistics, clustering (non)linear regression, non- and semi-parametric statistics tools are effective here. Some important general statistical issues are how to deal with high-dimensional noise, experimental design and multiple testing for combinations of different experimental neuroscience data platforms.

Biology-based Modeling and Methods

Development of dedicated, biology-based mathematical and computational models, and corresponding data analysis methods or predictive tools for neuroscience model systems with a smaller number of components are the focus of this theme. For instance, modeling and analysis of the connectivity structure and network dynamics based on observed spatio-temporal patterns in neuronal networks from multi-electrode arrays, and the investigation of the relationship between dendritic morphology and synaptic plasticity are part of the research in this theme, as are the development of models and analysis tools for the complex fluctuations in ongoing neuronal oscillations measured in EEGs. In a completely different area, population genetics, research like building models for assortative mating in extended twin families and considering the effects of cultural transmission are part of the theme. Tools that are used and explored include dynamic Bayesian networks, hidden Markov modeling, branching processes, parametric and non-parametric Bayes methods, self-organized criticality, and graphical models.

Integrative Data Analysis

This theme concerns the application of the integrative methods developed in the two other themes, i.e. the actual analysis of neuroscience data–in collaboration with one or more of the other research programs–within the themes of the Neuroscience Campus with the models and tools developed in the two other themes. Application of the methods may also point out limitations of current methods and may signal the novel directions in research methodology. This theme encompasses the integrated use of genome-wide association data (as can for example be obtained from public databases such as dbGaP), data available in UniGene databases, Ingenuity, and the systems-biology-of-the-synapse database of CNCR. Methods that are applied involve those for genomic imputation, gene-gene interactions, gene-network analysis, neuronal networks, and brain imaging. 

Hardware

Due to the scale of data that is generated by current genetic and neuroscience experiments, the use of cluster computers in genetics and neuroscience is of crucial importance. This requires knowledge of UNIX systems, shell scripting and parallelization. Part of the analyses mentioned above are carried out at the Genetic Cluster Computer (GCC). The GCC is hosted at the national Dutch computing facilities SARA and is financially supported by NWO and the VU (480-05-003, PI Posthuma). It consists of 64 dual core processors, adheres to strict security rules and is exclusively reserved for researchers working in the area of genetics, including neuroinformatics. Access to GCC also includes access to the broader Lisa cluster, currently totaling 800 computing nodes.

Executive Summary

The Integrative Analysis and Modeling program provides the essential expertise on modeling and data analysis for the Neuroscience Campus’ systems biology approach of the brain. It develops and assesses statistical methods and computational tools for the analysis of large, complex data sets. It also builds biology-based mathematical and computational models and designs model-based statistical tools for model systems with a limited number of components that are used to unravel generic principles of the involved biological networks. The program has three aims:

• to develop novel research methodology for integrative research concerning large and complex data sets (Methods for Large Data Sets);
• to perform research aimed at developing innovative tools for dedicated mathematical and computational modeling and statistical analysis (Biology-based Modeling and Methods);
• to perform integrative analysis of data obtained from different disciplines and levels of data (Integrative Data Analysis) 

Future perspectives

The program’s collaborative research projects concern modeling and analysis of data in the areas of genetic epidemiology, gene expression analysis and gene network modeling, phenomics, neuronal networks, and brain imaging. The research will result in mathematical and computational models, statistical methods and data analysis tools for biological networks, as well as high-level bioinformatics algorithms, including the integration and collaborative annotation of -omic data, and optimalization of statistical analyses in a cluster environment. It will also be instrumental in the transfer of technology and skills into ‘-omic’ bioinformatics, and in design of experiments and predictive biology.