Summary

The MOSAIC group develops and applies methods from computer science (data analysis, modelling, and simulation) to gain an understanding of the working mechanisms of living systems based on the first principles of physics.
 

Introduction

In order to understand the relationship between genotype and phenotype, it is necessary to consider both the spatial organization and compartmentalization of living systems as well as the fundamental physics of the interactions. This involves the automated (unbiased) processing of large amounts of experimental data. For phenotypic studies, these data are usually acquired in the form of digital videos or images. Methods from statistics and machine learning enable the identification of the physical phenomena and interactions that are present in a given system. Such hypothetical models need to be tested by comparing their behaviour to the real system by means of computer simulations. Computer simulations allow for the control and observation of all variables, which is not possible in classical experiments. Due to the complexity of biological systems, new computational methods for data analysis and large-scale simulations are needed.
 

Projects and Services

Data analysis and image processing

The MOSAIC group develops image analysis algorithms for a variety of biological and medical applications. A large class of these algorithms is based on a physical model of the imaged system. In recently developed algorithms, image processing involves optimal reconstruction of the system states using Particle Filtering. The algorithms are implemented in an open-source software framework that enables easy use and deployment. In addition to image processing, a novel sensitivity analysis method is being developed that provides parameter sensitivities (and their confidence intervals) for free during model fitting. Other data analysis activities in the group include feature selection methods for classification and pattern detection in trajectories

Modelling

The MOSAIC’s system models are resolved in space and time. They are based on the first principles of physics in order to provide certain predictive capabilities. This also enables the integration of phenotypic features with large datasets of genetic, physical, and metabolic interactions in order to identify the interplay among the functional modules and molecular machineries. Modelling techniques include partial differential equations, particle models, discrete state models, cellular automata, and multi-scale statistical models.

Computer simulation

The MOSAIC group studies model behaviour using fully resolved spatio-temporal simulations running on parallel high-performance computers. Currently used numerical techniques include adaptive hybrid particle-mesh methods, discrete element methods, lattice Boltzmann, and molecular dynamics. The group is actively involved in software engineering for large parallel computers and co-develops the Parallel Particle Mesh (PPM) library, a generic middleware that supports automatic and transparent parallelization of simulations. Besides enabling state-of-the-art performance without specific knowledge in parallel computing, the PPM library also reduces the code development time from years to weeks. The intention is to further develop the PPM library into an open-source simulation platform for spatio-temporal models in the life sciences.

Software Services 

• The PPM simulation library
• Image-J plug-ins for single-particle tracking, point detection, object outline segmentation, background removal, and 3D as well as 4D model-based tracking
• A pattern detection and segmentation tool for trajectories
• Simulation tools for geometry-corrected evaluation of quantitative FRAP experiments in complex-shaped organelles
• Tool for statistical motion analysis based on moment scaling spectra

Websites for Further Information

MOSAIC group: http://www.mosaic.ethz.ch/