WWL: Current Implementations for Research

The above examples demonstrate the use of the world wide laboratory architecture for service, collaboration and education. Using this architecture for remote scientific research where reliance on a local instrument operator is minimal or non-existent poses additional requirements which will be discussed below:

Automated/Intelligent Control for Scientific Research

Our group has been extensively involved in a project involving scientific research on remote and automated instrumentation. The goal of the project is to acquire very large numbers of good quality images from a transmission electron microscope (TEM) completely unattended by a human operator. The motivation for developing this automated system arises from the field of electron crystallography in which TEM is used to study the structure of proteins at moderate to high resolution (5 to 30 A). The technique most commonly used to preserve the proteins in the TEM is known as cryoEM in which the protein is preserved in a very thin layer of vitreous ice [9]. The ice is usually suspended over a holey carbon grid and the goal of the microscopist is to identify holes where the ice is potentially of the right thickness and acquire a high magnification image of this area (see fig. 4).

Figure 4: Acquisition of cryo-electron micrographs. A copper grid (a) is covered with a carbon coated holey plastic mesh (b). A droplet of buffer containing the protein of interest is applied to the grid, blotted to a thin film, and then rapidly plunged into a liquid cryogen. The protein of interest (c) is preserved in its native form in the vitreous ice.
Figure 4

There are a number of practical problems with the CryoEM procedure that tend to make it extremely time consuming and tedious for the operator. The first is that producing ice of precisely the right thickness is not straightforward, as a result the ice is quite often either too thick or too thin and this entails a lot of searching around the grid to find suitable areas. Secondly, because the electron beam is extremely damaging to the specimen and will destroy it after a very short exposure, the grid can only be examined for any length of time at very low magnification. The high magnification image is never examined prior to shooting the micrograph and not too surprisingly this leads to a rather high rejection rate; many of the acquired images are simply thrown away and only a few turn out to be suitable for further analysis. Finally, because the beam damages the specimen the micrographs must be acquired with a very small dose of electrons and as a result the images are very noisy. Thus, to properly determine the protein structure to high resolution the signal to noise of the structure must be increased and this requires averaging together many images. The end result is that this technique by its nature requires the acquisition of large numbers of micrographs, perhaps thousands, or tens of thousands to achieve high resolution. Manual methods are clearly impractical and it was with this in mind that we embarked upon the project of completely automating the acquisition of large numbers cryo-electron micrographs.

As a prototype for automated TEM acquisition, we have developed a system, called Leginon [10] to automatically acquire large numbers of acceptable quality images from specimens of negatively stained catalase, a biological protein that forms crystals. Acquiring good quality images of this specimen is often used as a test for students taking a course in electron microscopy and thus provides an excellent driver for the research methods that must be developed to solve the general problems of automated image acquisition. Furthermore, as catalase is an ordered crystalline structure, assessment of this order provides us with an objective measure of the quality of the automatically acquired images (see fig. 5). Each low magnification image (fig. 5a) is processed to identify large contiguous areas of density by a template matching method. Image feature metrics (size, mean, variance, centroid) are calculated and stored for each of the identified contiguous regions. These image features are later used in deciding whether a high magnification image (fig. 5b) of the region will be acquired; for example, regions that are too small are rejected. The image quality of each high magnification image is automatically assessed by calculating the power spectrum (fig. 5c), identifying diffraction spots (fig. 5d), and measuring the signal to noise ratio of each diffraction spot.

Figure 5: Automated acquisition of electron micrographs of negatively stained catalase crystals.
Figure 5a Figure 5b
a b
Figure 5c Figure 5d
c d

Currently, the automated system can acquire approximately 1000 images in a 24 hour period. In one experiment, we have compared the performance of the automated system to that of a human operator. A total of 288 high magnification images were acquired manually and 79% of these were acceptable as defined by an analysis of the order of the crystal. In comparison, using the same specimen, the fully automated image acquisition system was used to acquire 380 images of which 51% were acceptable.

The system can be further improved by adding intelligence to the feature selection criteria. For example, analysis of the results indicated a correlation between average feature intensity and image quality. This feature intensity is related to the thickness of the catalase crystal and indicates that thinner specimens result in more acceptable images. The fully automated target selection criteria was further refined by incorporating an assessment of specimen thickness into the model. By acquiring high magnification images of only those features that have an average intensity greater than a preset threshold the percentage of acceptable images can be significantly improved. For example, if the threshold is set to 6000, the percentage of acceptable images improves to 86% from a baseline of 51%. Thus, the automated system does as well or slightly better than a human operator.