IIMS Workshop Report

Presentations

J.M. Carazo, CNB-CSIC, Madrid, gave an introduction to the IIMS project and its predecessor BioImage.

Kim Henrick (EBI) [HTML] [PPT] gave an introduction to data harvesting and the need for a common language for information exchange. Related work being carried out for NMR, and X-ray crystallographic techniques was shown together with related work on protein production and high throughput automation work being carried out within the EBI.
Some related previous work on standards is given in

3D cryo-EM data resolution criteria - Prof Marin van Heel, Imperial College, London presented analysis of resolution definitions for 3D-EM. He showed that the concept of splitting the (3D) data set into two halves and obtaining the Fourier Shell Correlation (FSC) at a fixed '0.5' threshold cannot, for example, account for the varying number of voxels in a Fourier 'shell' and that this threshold criterion should be considered a step back from the 2-sigma threshold criterion, originally introduced by Saxton and Baumeister (1982). The 3-sigma threshold (three standard deviations over the random noise value) must be multiplied by a number of factors including the square root of the number of asymmetric units within the given point group symmetry. The reason for using this correction factor is that the number of independent voxels in the two Fourier-space 3D shells one compares is lower for symmetrical particles than it is for asymmetrical ones. The 3-sigma criterion indicates at which resolution level one has started to collect information significantly above the expected noise levels. The new 1 bit and 1/2 bit threshold criteria indicate at which resolution one has already collected enough information of a reliable interpretation of the map. He further argued that any resolution claim should remain under approximately two-thirds the Nyquist frequency. In summary Marin believes that use of any fixed-value threshold criterion like the 0.5 FSC criterion, is essentially wrong, especially when used in connection with asymmetric particles should not be used. A manuscript describing these ideas is in preparation.
Saxton and Baumeister, J.of.Micr., 127,(1982) 127-138;
M. van Heel, Ultramicroscopy., 21,(1987) 95-100;
Orlova, et.al., J. Mol. Biol., 271, (1997) 417-437.

NCMI - EM project management and data storage system - Dr Steven Ludtke, National Center for Macromolecular Imaging Baylor [HTML] [PPT] described an object-oriented database for internal record keeping and the use of an 'electronic notebook' approach to keeping experimental data. This generated approximately a tera-byte of archive data each year. Binary data was stored as external files on a RAID. He also described the ease of use and flexibility of this type of data-base compared with a relational one in its ability to integrate experimental data with the 'EMAN' reconstruction software and allows data mining. Data collection involved loading tabs from the machines, eg the 'vitri-bot' and acquiring micrographs from the CCD camera, using Zope (Zope is a leading open source application server, specialising in content management, portals, and custom applications) and written in Python.

William Baxter, RVBC Wadsworth [HTML] [PPT] gave a review on the RVBC data management project encompassing Automated Microscopy, a Project Archive and a Reconstruction Engine was described. The automation work on microscopy tackles the task of experiment conditions and of controlling the writing of micrographs with a buffer computer and a pipeline of working directories, at the same time recording microscope parameters in the Project Archive and signaling the reconstruction procedures to begin processing. Everything is project based and spider batch files, and their input and outputs are recorded. Standard data format is not emphasised, the Archive Database is used to store common data. However the system could be modified with a deposition button to send XML to the emdep deposition system.

Standard Conventions of Data Exchange & Archiving - Dr David Belnap, NIH, Maryland [PPT] [PDF] described Standardised Conventions for Data Interchange Among the 3D EM Community. He further described a tag-based image file format for electron microscopy. cf Appendix

Dr Reiner Hegerl, Max-Planck-Institut fuer Biochemie, gave a presentation on the problem of de-noising 3D-EM images. Many data sets collected using electron tomography techniques suffer from a poor signal-to-noise ratio, which severely hinders the application of methods for automated data analysis, such as feature extraction, segmentation, and visualization. Limitations arise from the restricted tilt range, different contrast contributions, radiation-induced object modifications, and a high noise level. A denoising algorithm that is suitable for our application must be able to preserve as much as possible of the signal while reducing the noise to a sufficiently low level. Dr Hegerl presented a multidimensional implementation of two powerful signal reconstruction techniques, namely invariant wavelet filtering and nonlinear anisotropic diffusion. The appropriate multidimensional implementations of wavelet and diffusion techniques allow for a superior performance over conventional noise-reduction methods. Extending the 1D translation-invariant denoising algorithm of Coifman and Donoho to higher dimensions proves to overcome the poor performance of orthogonal wavelet transforms. A method to quantify the loss of information due to denoising artifacts on data with an unknown signalnoise relationship, and a scheme for denoising of such data was outlined.

Ansgar Philippsen, Muller Inst, Basel [PPT] described an Image Processing Toolbox Project written in Cobra and Python. The url giving latest details of this project is: IPT (cobra.mih.unibas.ch/ipt/). He also mentioned DINO (cobra.mih.unibas.ch/dino/). DINO is a realtime 3D visualization program for structural biology data.It runs under X-Windows and uses OpenGL. Supported architectures are IRIX, Linux (i386 and PPC), OSF1 and SunOS. DINO is distributed in binary form only, the current DINO version is 0.8.4. Structural Biology is a multidisciplinary research area, including x-ray crystallography, structural NMR, electron microscopy, atomic-force microscopy and bioinformatics (molecular dynamics, structure predictions, surface calculations etc). The data produced by these different research areas is very diverse: atomic coordinates (models and predictions), electron density maps, surface topographs, trajectories, molecular surfaces, electrostatic potentials, sequence alignements etc. DINO aims to visualize all this structural data in a single program and to allow the user to explore relationships between the data. There are five data-types supported: structure (atomic coordinates and trajectories), surface (molecular surfaces), scalar fields (electron densities and electrostatic potentials), topographs (surface topography scans)and geom (geometric primitives such as lines). The number and size of the data the program can handle is only limited by the amount of RAM present in the system. No artifical limits are set. Supported input file formats are PDB (coordinates), X-PLOR/CNS (coordinates, electron densities and trajectories), CHARMM (coordinates, trajectories and scalar fields), CCP4 (electron densities), UHBD (electrostatic potentials), DELPHI/GRASP (electrostatic potentials), MSMS (surfaces), MSP (surfaces), GRASP (surfaces), MEAD (coordinates and electrostatic potentials) and greyscale TIFF (surface topographs). Output can be written as PNG, PostScript and POV-Ray (v3.1 and v3.5).

Richard Newman, EBI [HTML] [PPT] gave an introduction to the data model for electron microscopy developed within the IIMS project. The PDB template for EM-maps containing fitted coordinates, the data model for the description of the basic EM experiment and the XML descriptors was introduced. The plan to release the deposited data sets via an FTP area was presented.

Mohammed Tagari, EBI [HTML] [PPT] outlined the emdep deposition system and the XML, Java servlet and tomcat technologies used were described. The current system was mostly hand-written a new system has been developed that is dictionary driven through XML templates was described. The UML technologies used in the EBI for the protein production LIM system will be used in the next emdep system.

MaryAnn Martone, UCSD, gave an overview on the design, implimentation and access to the Cell Centered Database (/ncmir.ucsd.edu/CCDB/) was presented along with the tools for integration of multi-resolution and multi-disciplinary neuroscience data. The CCDB is part of the Biomedical Informatics Research Network (BIRN) and contains data sets derived from 3D light and electron microscopic reconstruction techniques, particularly electron tomography. The CCDB models the entire process of reconstruction, from specimen preparation to segmentation and analysis. A volume reconstruction is stored along with pointers to all of the raw images and the processing details required to reconstruct the volume from the raw data. The CCDB integrates oracle 9i and SRB (the Storage Resource Broker) to manage 2D or 3D microscopic imaging data. The data sources are linked to other information sources using a mediation-based approach. The first ontology to be incorporated into the mediator will be UMLS, with the Neuronames extension linking anatomical concepts in the database to the relevant ontology IDs in UMLS. Additional knowledge bases will be added as the project progresses, and participating groups will be able to define their own ontologies and incorporate them into the BIRN. An introduction to the KNOW-ME (KNOWledge-Map-Explorer) system for ontology semantic browsing of the CCDB integrated data using knowledge maps was given. The concepts of, Knowledge Maps, semantic browsing for concepts and processes was also shown.

Professor Mark Ellisman, UCSD, reviewed the concept of Telescience/Telemicroscopy was presented as remote access for data acquisition and analysis with a web and grid based system using cross platform Java for management and access control. The system is centred on neuroimaging and brain chemistry with a heterogenous collection of data from the complex structure of the microscale domain to the ultrastructure of the nervous system. The newly created Biomedical Informatics Research Network (BIRN), is an NCRR (www.ncrr.nih.gov) initiative aimed at creating a testbed to address biomedical researchers' need to access and analyse data at a variety of levels of aggregation located at diverse sites throughout the country. Specific BIRN program objectives include:

• establish a stable high-performance infrastructure linking key NCRR-supported P41 and GCRC sites using the Internet 2 network
• establish distributed and linked data collections for the testbed projects
• enable the use of distributed, heterogeneous, grid-based computing resources for project-specific data storage and collaborative analysis
• enable data mining from multiple data collections or databases on neuroimaging
• develop stable software and hardware infrastructure that can be reapplied and/or expanded to include other sites with different research foci
• demonstrate effectiveness of these technologies in improving and extending research results (needs pull not technology push)

BIRN has three types of data:

• Data from devices or analysis, such as images and volumes.
• Metadata describing that data, like the name of the researcher that collected the data, and when they did it.
• Domain atlases that categorize and sub-categorize data, like the parts of a brain or a cell.

From micrograph digitisation and particle selection to 3D map computation with a C package under Linux OS (on a PC) - Prof Salvatore Lanzavecchia, Universita' degli Studi, Milan Salvatore Lanzavecchia described how he had developed algorithms for 'unbending' and orientation of pairs of particles. He also described mapping conventions and the creation of euler angles derived from projection algorithms and how easy to switch from one system to another. He used helical reconstructions of tubules to demonstrate the use of helical unbending. Programs were written in C and were available on personal request.

Professor Lars-Goran Ofverstedt, CMB, Karolinska Institutet, outlined the results of using the software COMET, constrained maximum entropy tomography. used with Electron Tomography (ET), to improve the quality of imaging data of 2D and 3D images. Resolution and noise reduction are key factors in imaging. When analyzing data from experiments performed by electron microscopy (EM), ET or similar methods, the smallest detail that can be reproduced is limited by the resolution of the technique. An iterative algorithm is used to generate high fidelity reconstruction of the observed object using a stable regularisation method using low-dose electron tomography (ET) allowing visualization of individual objects in the sample.

Dr Andreas Hoenger, EMBL [HTML] [PPT] used a study of kinesin and other molecular motors by Helical and tomographic 3D reconstructions of microtubules to indicate the complexity of presenting these data to a wider audience. It was clear that for an archive that certain types of cryo-3D EM data required a family of results to give the full understanding of the particular experiment.

Dr Henning Stahlberg, Biozentrum, used the family of Aquaporins as an example of structure determination by using EM and AFM Studies on 2D Crystals. The two complementary techniques - electron microscopy and atomic force microscopy (AFM) - are employed to study 2D crystals of membrane proteins, in particular members of the aquaporin superfamily. Atomic models of aquaporins - based on EM and x-ray data - have been reported and the efforts being made to increase the speed of structure determination from 2D crystals were presented. The techniques of X-ray diffraction on single crystals and 2D electron diffraction are complimentary and this family of proteins clearly show that the key features of the C-terminal residues are only resolved and visualised by AFM. The combination of 2D crystallization, electron crystallography and AFM for the study of membrane proteins is attractive for the following reasons: First, the membrane protein is embedded in its native environment and is thus more likely to remain fully active. Second, obtaining 2D crystals of a membrane protein is less difficult than 3D crystals. Third, the dynamics of the membrane protein surface can be assessed under physiological conditions by AFM.

Professor N Volkmann, Burnham Institute, CA, presented a study on atom fitting to EM maps. Recently, algorithms were developed that allow objective and automatic combination of atomic resolution structures with lower resolution EM reconstructions (Jiang et al., 2001; Roseman, 2000; Rossmann, 2000; Volkmann et al., 2000; Wriggers and Chacon, 2001). These algorithms allow building atomic models of large complexes imaged by EM using the atomic structures of their components. Dr Volkmann explored the potential usefulness of such docking studies in the context of cellular electron tomograms. Specimen holders restrict the possible tilt angles of the specimen to about ±70º for cryogenic data collection, leading to a missing wedge of data. As a consequence, features perpendicular to the electron beam are better resolved than features parallel to the beam, resulting in anisotropic resolution and distortions of the structure. Here, we examine the influence of limited resolution, low SNR and missing wedges on docking of atomic models into calculated electron tomograms. A refinement procedure based on a density correlation-like measure was used successfully to improve the fit of pre-built models. A program system COAN (Volkmann and Hanein, 1999) to evaluate docking of atomic models into calculated electron tomograms was outlined. The calculations indicate that even with the current limitations in resolution (50-100Å) and quality of electron tomograms a meaningful docking that gives near atomic or residue-level information may well be achievable, especially if the docking is combined with advanced segmentation algorithms. While the SNR and the resolution seem to be of no major concern, the artifacts introduced by the missing wedge problem appear to be a more serious limitation. This problem may in part be solved by more complete data collection strategies such as double-tilt tomography.

Professor Wah Chiu,National Center for Macromolecular Imaging, Houston, TX [HTML] [PPT] detailed studies from electron cryomicroscopy where it is possible to reconstruct single macromolecular machine with high symmetry such as icosahedral virus to an intermediate resolution of 7-9 Å. At this resolution, one can expect to visualize helices and beta sheets of protein components. His group’s software such as helixhunter has been developed to help identifying the locations of helices of protein components automatically. Though the density map at this intermediate resolution cannot resolve individual beta strands, the beta sheet can be recognised as a continuous and flat density. Additional information is provided through bioinformatics analysis. Sequence based analysis can be employed to predict the sequential order and lengths of the secondary structural elements. This information can be used to match with the locations and the length of helices and the sizes of beta sheets observed in the 3-dimensional map. These matches can result in the assignment of amino acid residues to the observed secondary structure elements and thus yield models of protein folds in different domains.

Helen Saibil, Birkbeck College, London, gave a presentation on the problems of an end-user and attempts at atomic structure fitting to intermediate resolution cryo EM maps. The overall conclusion was that currently manual fitting of PDB chains into an EM map was more successful that several automated procedures.