Log In Join
 |  Forums  |  Help  |  Settings  |  Download


Research: Help Conquer Cancer: Project FAQs
 
Project Overview
Project Details
Project News
Research Participants
Project FAQs
Project Statistics
Help Conquer Cancer



There are several direct and indirect benefits of the project. For the first time, scientists will execute a comprehensive image analysis and classification of crystallography images. This will lead to better understanding of the crystallization process, and will enable scientists to improve the accuracy and speed of CrystalVision. Improved understanding of the crystallization process and improved CrystalVision also will enable more disease proteins to be crystallized faster. Finally, more 3D structures will improve our understanding of disease and potentially its treatment, and will lead to improved in silico (performed on a computer or via computer simulation) structure prediction.

On the lowest level, CrystalVision will compute thousands of image features for each crystallography image. This data objectively measures characteristics of the image, which will enable scientists to use a system to discern image classification. In turn, this will allow them to automatically and objectively characterize results from the high-throughput crystallization screens, and then apply data mining techniques to optimize future crystallization experiments.

After careful analysis, evaluation and interpretation, all results will be published in the public domain. The scientists' first goal is to improve the CrystalVision system to enable automated, accurate and fast crystallography image classification. This algorithm will then be deployed at Hauptman-Woodward Medical Research Institute to ensure that this public high-throughput crystallography screening facility will speed up crystallization of many disease-related proteins.

Each work unit is a photograph of a protein crystallization experiment (one out of 1,536 images per protein, photographed six times over a period of one month), a visual record of the state of a protein sample dissolved in a solution of crystallizing agents. This photograph is shown in the background of the agent window. The Grid agent performs a computer vision analysis of the image in order to interpret its contents, first determining important image features, which are then used to classify (or label) the result of the experiment. During the feature image computation, intermediate steps of this analysis are displayed in the colored circles appearing in the foreground of the agent window.

The analysis is a search for four large categories of features in the image: microcrystals, straight lines, discrete objects, and textural features. Intermediate steps of the texture analysis are displayed in the colored circles that appear in the foreground of the agent window. As each step is completed, the computed result appears in the agent window. Each circle is a copy of a region of the original image, transformed to highlight a different texture.

The background image is a photomicrograph of a protein crystallization experiment. The experiment takes place in a droplet of water the size of a pinhead (200 nl), suspended in an oil-filled chamber. The circular wall of the chamber, and the roughly circular droplet contained within are visible in the photo. Inside the droplet, precipitated protein or salt, or even protein crystal may be visible.

Each disk is a visualization of a different texture measure applied to the background image. Thus, when two disks are differently colored, it means only that different textures are more or less prominent in different regions of the image. Twenty-six measures of texture are visualized in the Grid agent.

Each measure is related to frequencies of the grey-scale values of pairs of pixels found in the image, and summarizes these frequencies according to pixel-pixel contrast, correlation, variance, or entropy. Each of 13 categories of statistics is measured multiple times by changing the distance and relative orientation of the pixel-pairs.

Each disk visualizes the results of a search for a particular texture in the original image. The texture search is done in three steps. The first step records fine-grained changes in the grey-tones of the image, the second step records medium-grained changes, and the third step records coarse-grained changes. The three steps are visualized together by using red (step 1), green (step 2), and blue (step 3) colour channels to create a full-colour image representing the whole process. A blue region of the disk would then indicate a region of the original image where the texture is most apparent in coarse-grained grey-tone changes.


The Grid agent will only display the results of the last 10 image analysis steps. As the next step is completed, its result is displayed, and the oldest is removed.

This failure occurs for work units running on the graphics card during a test of the ability of your graphics card to run the work unit. If you see this error, make sure you are using the latest video driver for your graphics card. Also, remote desktop connections can have an effect on graphic card applications and it is recommended not to have an active remote desktop connection while the Help Conquer Cancer graphics card application is running.

You may learn more about updating your video drivers at the website of your graphics card manufacturer: Nvidia or AMD

At the beginning of each work unit run on your graphics card, a small portion of the workload is run to estimate the execution time of a single kernel execution on the graphics card. If this estimate is too high, the application will exit to reduce the risk of Windows restarting the display driver due to the Timeout Detection and Recovery feature of Windows. If this occurs, the above error message will be written to the stderr log. If this occurs multiple times, it is likely the graphics card is not capable of running the project. Please refer to the "What graphics cards are not able to participate in the Help Conquer Cancer research project?"FAQ for a list of graphics cards which are not supported.

If it occurs occasionally but not on every execution, it could be that other graphics intensive work is interfering. We recommend that you set your preferences to not allow World Community Grid to run while you are actively utilizing your computer. This option is available on the Device Profile page under the custom options section. This option is labeled "Do work on my graphics card while computer is in use?". Select "no" and save.

The Help Conquer Cancer graphics card application requires the OpenCL extension cl_khr_local_int32_base_atomics and will not run on cards that do not support this extension. If you see the error above it is because your graphics card does not support the extension.