ÀÖÌìÌÃÓÎÏ·ÍøÕ¾

PDBe’s new high-performance graphics workstation supports the team’s efforts to explore and annotate very large, 3D electron microscopy datasets. The acquisition was motivated by practical challenges curators face when trying to zoom in on details of large bioimaging datasets, which can be millions or even billions of datapoints in size. In particular, the team had a real need for better tools to check and annotate segmentations of 3D cellular electron microscopy and tomography data.

The PDBe team is currently developing EMDB-SFF, a format for segmentation data, based on input from a working group of international community experts. PDBe plans to develop tools to annotate these segmentations with ontology terms and identifiers from key biomolecular archives.

The investment was made possible by a capital grant for image data repositories coordinated by Jason Swedlow (Dundee), in which the University of Dundee, the University of Cambridge and ÀÖÌìÌÃÓÎÏ·ÍøÕ¾ work together.

"A beast of a machine"

For the geeks, the workstation is an HP Z840 with Intel Xeon 2.5 GHz 12-core, 30 MB L2 CPU, 32 GB DDR memory, 512 GB PCIe SSD and 2 TB SATA SSD storage. The graphics are produced by an NVIDIA Quadro M4000 8 GB card with 1664 CUDA cores, and displayed on a Wacom Cintiq Touch 27� QHD Touch Graphics Tablet.

once calculated that, forty years ago, a graphics workstation useful for structural biologists cost the , but the cost has now come down to about 5% of one Ferrari.

The graphics display in action (the structure shown is that of HIV1, with 13 million atoms currently the largest in the PDB).

PDBe’s Paul Korir assembled the workstation from parts. About its performance potential he says: “Despite its breathtaking performance it still has a lot more room for expansion. For example, it can accommodate a second Intel Xeon CPU with up to 22 cores, memory of up to 1 TB, up to 10 internal drive bays in various RAID configurations, support the highest data transfer over 10 Gigabit Ethernet as well as host a second state-of-the-art graphics card. It's a beast of a machine.�

Paul Korir busy assembling the high-performance graphics workstation.

More than pretty pictures

Once the workstation had been assembled, it didn’t take long for members of the PDBe team to discover several other applications that will make their lives a lot easier. Maya Holmdahl is a trainee from the University of Uppsala (Sweden) who is involved in analysing the results of the EMDataBank Map Challenge. She says: “This machine enables me to work with large data sets. It makes it possible to manipulate EM maps containing a huge amount of information interactively.�

The PDBe annotators, who curate structure depositions to PDB and EMDB, have found that the workstation allows them to work with very large structures and datasets that would bring their regular desktop computers to a grinding halt. PDB and EMDB annotator Alice Clark says: “We are seeing a large increase in the number and variation of entries being deposited to EMDB. In addition, the annotation of EMDB data is under development. This workstation is a great tool for our annotation work, but also for the public engagement activities we are involved in here at PDBe.�

As alluded to by Alice, another obvious application of the workstation is in producing the kind of stunning 3D graphics images that adorn scientific publications, posters, brochures, etc.

Ardan Padwardhan is leading the MOL2CELL project, which also includes the EMPIAR archive of raw electron-microscopy image data. “This workstation enables us to work with real, hard-core 3D reconstruction and segmentation data in a range of software packages such as Amira, IMOD and Chimera, thereby greatly facilitating the development of web services and tools to support this data in EMDB and EMPIAR.� One of the next steps will be to explore ways to represent and visualise results from correlative light and electron microscopy (CLEM).

The images below illustrate some of the applications mentioned above.

Example of an EMDB map automatically segmented using the program  .

A segmented image of a  displayed with the program  . The original dataset contained 15 billion elements (voxels). Data kindly provided by  .

·¡²Ñµþ³¢-·¡µþ±õ’s&²Ô²ú²õ±è; and  receive a demonstration of the workstation from PDBe’s Paul Korir (left) and Alice Clark (right). Alvis uses the pinch movement (familiar from phones and tablets) on the touch-sensitive screen to zoom in on the displayed objects, a tomogram and a cartoon rendering of the HIV1 structure (consisting of 13 million atoms).