Description: Application of machine learning algorithms for clinical predictive modeling: a data-mining approach in SCT Bone Marrow Transplantation 49, 332 (March 2014). doi:10.1038/bmt.2013.146 Authors: R Shouval, O Bondi, H Mishan, A Shimoni, R Unger & A Nagler

Description: In myocytes, small heat shock proteins (sHSPs) are preferentially translocated under stress to the sarcomeres. The functional implications of this translocation are poorly understood. We show here that HSP27 and αB-crystallin associated with immunoglobulin-like (Ig) domain-containing regions, but not the disordered PEVK domain (titin region rich in proline, glutamate, valine, and lysine), of the titin springs. In sarcomeres, sHSP binding to titin was actin filament independent and promoted by factors that increased titin Ig unfolding, including sarcomere stretch and the expression of stiff titin isoforms. Titin spring elements behaved predominantly as monomers in vitro. However, unfolded Ig segments aggregated, preferentially under acidic conditions, and αB-crystallin prevented this aggregation. Disordered regions did not aggregate. Promoting titin Ig unfolding in cardiomyocytes caused elevated stiffness under acidic stress, but HSP27 or αB-crystallin suppressed this stiffening. In diseased human muscle and heart, both sHSPs associated with the titin springs, in contrast to the cytosolic/Z-disk localization seen in healthy muscle/heart. We conclude that aggregation of unfolded titin Ig domains stiffens myocytes and that sHSPs translocate to these domains to prevent this aggregation.

Description: Aim of this project is to develop novel methods for a better understanding of temporal signatures - both in their spatial and spectral domains - arising from multiple components (subsystems) of a Geodetic Earth System model (GESM). This model is specifically dedicated to predicting geodetic measurements, either coupled or independent for each subsystem. Methods of interactive visualization provide the means to subsequently explore variations of mass signals over time in the spatial and spectral domains, assess their superpositions in the geodetic measurements, and identify their signal and noise characteristics. This information is important to assess the sensitivity of the geodetic measurement to a specific processes, and to explore methods for separating superimposed signals. In this paper, we present requirement on the system’s design and the visual interface to support exploration of mass signals over time in the spatial and temporal domain. As a first result of our research, we propose a data model that supports the retrieval of all relevant spatial and temporal aspects of the data in real-time.

Description: The geological record of past sea levels is unevenly distributed, both temporally and spatially. Further interrogation of this data (e.g. for geophysical or glacial-isostatic modeling purpose) necessitates compiling data from literature or data repositories and amounts to the collection of thousands of individual sea-level index points. The spatiotemporal context is often lost during this process. The heterogeneity of the data with respect to significance and content also prevents the application of many statistical techniques. A solution to this problem is the application of advanced visualization concepts, which maintain the connection between data points and associated metadata (e.g. location, dating control, type of sample, etc.) during modeling. This allows fore a more complete overview and communication of sea-level data and modeling output. We present the visualization framework, SLIVISU, developed by GFZ-Potsdam, which provides multiple linked views including synoptic analyses of the geological data, model configurations and predictions. SLIVISU is applied to variations in past sea levels during the last glacial cycles using geological and archaeological data.

Description: In many scientific environments, scientists face two major difficulties in observing spatial and temporal dynamics of geo processes. First, measurements of an observable usually contain a superposition of signals of interest; the interpretation of a measurement therefore requires scientists to simultaneously assess multiple complementary observables and/or simulate the processes generating the signals. Second, many quantities of interest cannot be directly measured but need to be inferred from a limited amount of observables. Here we present a prototype visualization tool for supporting scientists in exploring the spatial and temporal mass change variations within System Earth. Mass change cannot be measured directly but need to be derived from satellite measurements of the Earth’s gravity field. The first visualization shows the simulated mass anomaly (here, with respect to the annual mean), and the second visualization shows the calculated gravity field taking the simulated mass change as input data. The third visualization shows the mass change inverted from the measured gravity field of the Earth as input that is shown in the fourth visualization. Individual components of the simulated mass change arising from different subsystems can be (de)selected and the imprint in the gravity field is visualized. The user can navigate through time, as well as zooming and rotating of the spheres and adjustment of the color scale are possible. The behavior of all four spheres is synchronized. The combined representations of these visual views supports scientists 1) assess the overall quality of the modeled versus the observed gravity field signals, 2) determine the influence of different processes to the measured variation of the Earth’s gravity field in the region of interest, and to understand which fine-grained variation in the mass change cannot be inferred from the satellite measurements. This presentation will present the prototype visualization tool and provide a discussion on how our it supports geoscientists to study the relationship between simulated and mass change variation recovered from the Earth’s gravity field.

Description: Geoscientists collect various data to study system Earth. To extract all the knowledge inherent in the data is a challenge geoscientists have to face. Methods for data mining and visual data exploration developed in computer science facilitate knowledge extraction from data. Although these methods are mostly applied to business data, they also offer potential to extract knowledge from geoscientific data. In our research we adapt and enhance methods from data mining and visual data exploration to geoscientific requirements. Two examples are given that show how the novel methods contribute to geoscientific research: The extraction of interesting spatiotemporal patterns from large data sets and the comparison of data from simulation models and real world observations.