One of the largest challenges in obtaining X-ray diffraction data from biological samples is growing large, high quality crystals. Currently, there is not way to reliably predict successful crystallization conditions based on protein sequence alone and so high-throughput approaches are very appealing. High-throughput crystallization screens test a large chemical space using hundreds of different crystallization cocktails at the nano-drop scale. Successful conditions can then be scaled up and optimized to grow larger crystals.
The High-Throughput Crystallization Screening Center at the Hauptman-Woodward Medical Research Institute provides this high-throughput screening service to users, offering 1536 condition screens for both soluble and membrane protein samples. Each plate is imaged over a period of two months using brightfield microscopy, as well as SHG and UV-TPEF microscopy, using Formulatrix Rock Imagers.
This high-throughput produces a large volume of images that must be sorted through in order to pick out the best condition; a task that can be very tedious and repetitious.
In 2019 Bruno et al published Classification of crystallization outcomes using deep convolution neural networks which included a CNN model that could accurately classify crystallization screening images, opening the door to automating this process. The MARCO model is open-source and available for users to implement, and has been incorporated into See3 software for Collaborative Crystallization Centre (Australia) users, but has not been available in an average-user oriented graphical user interface (GUI) program.
Polo is therefore designed to incorporate the benefits of the MARCO model and integrate the functionality of established crystallization image labeling software such as MacroscopeJ to create a GUI targeted for HWI Crystallization Center users and others with crystallization screening experiments. Polo incorporates all the tools needed to go from raw crystallization images to designing optimization screens without the need to install any dependencies.
For more information, please visit the documentation page linked below.
Download the latest release here. A self-installing version is available for Windows 10. Executable files are available for Mac OS >= High Sierra and Ubuntu >= 18. Please see the installation help page of the Polo website for solutions to common issues users have encoutered.
If you encounter a problem that is undocumented, please let us know either through the bug reports and suggestions form or by opening an issue in this repository.
We have created a number of reasources to help you get going with Polo. If you prefer textual guides you can visit the User's Guide page of the Polo website. If you are more of a video tutorial person check out the Video Guides page where Ethan will walk you through how to use some of Polo's primary interfaces.
Polo strives to be accessible, easy to use and reduce human work. If you have a suggestion of how any of these or other aspects of the program could be improved please make it known! You can do so through the suggestion form or by opening an issue on this repository. Pull requests encouraged!
View your images serially, in a slideshow like format, with or without running the MARCO model. Sort images by their human or MARCO classification and/or classification confidence. Imaging runs of the same sample are automatically linked allowing for easy creation of time-resolved views. Toggle UV-TPEF or SHG (or both) images when available with one click to confirm the presence of crystals in a well.
You can also view many images at once using the Plate Viewer which helps with rapid initial screening of images. Highlight images within a grid by their human or MARCO classification to make confirmed or potential crystal hits stand out. Select imaging images to open them in a zoomed in viewer.
Polo allows you to share your results with other humans or machines easily. Export images of a specific classification along with their cocktail and plate data to a ready-to-present PowerPoint (pptx) like the one shown above or to machine parsable formats like csv or json.