MARCO Polo: Finding Crystals

Published September 30, 2020 This content is archived.

This is an image of Ethan Hollerman's MARCO POLO poster with diagram showing workflows, graphic user interface, sharing results, and accessibility across various elements of the program. The intro saus a m,ajor step towards the automation of this process was taken in 2018 with Bruce et al reporting the Machine Recognition of Crystallization OUtcomes model, a deep convolutional neural network capable of classifying crystallization images with over 90% accuracy. While the model has found use in commercial level closed-source software, it has remained largely inaccessible to averageusers. Our GUI, Polo, seeks to fill this gap by providing one-click use of the MARCO model with intuitve imae viewing and classification interfaces that is frere -to-us and mostly for bot academic and commersial use under the terms of the permissive GPL-30 softare license. Workflows shows how a typical use might incirporate polo into their high-throughput crystallization workflow. Screening images as downloaded directly from the hight throughput crystalization screening cnter via Polo's integrated FTP client, the user then uses Polo to classify and review their screening images for the hits and can share their results to both Polo users and non-Polo users using a variety of file formats. THE GUI shows screenshots of the slideshow view interface displaying brightfield images of Wel Umber 400 of a lysosymesample set u at the HWI Crystallization Center. Image details and cocktail details are showin in the inset to the right. Using the Snap Spectrum button will show UV-TPEF and SHG imaging modalities if available. In adddition, a single-image views Polo includes a Plate Viewer interface that allows for viewing up to 96 screenings images at once. Images can be colored bu their human or MARCO classification and selected to be shown in a larger pop-out view provides functions equivalent to the Slideshow View.

Summer intern Ethan Hollerman won a Student Poster Prize at the 2020 LCLS/SSRL Users’ Meeting for his work on implementing the MARCO algorithm into a new GUI!  MARCO POLO is a multi-platform open-source Python-based graphical user interface that has been developed to provide access to MARCO automated classification and data management tools for biomolecular crystallization screening. MARCO is an academic and industry collaboration using image data from the Crystallization Center, Bristol-Myers Squibb, GalaxoSmithKline, Merck, and the Collaborative Crystallization Centre in Australia. It involved image analysis experts at Google Brain, The University at Buffalo, Duke University, and the University of York in the UK, with industry input from Formulatrix. MARCO is being actively used in industrial settings, in Formulatrix instruments, and soon for users of the Crystallization Center.  The GUI is undergoing beta testing, a paper is forthcoming, and the GUI will be released shortly.