Autoscoring of Protein Crystallization Drops in ROCK MAKER using MARCO
Aggarwal, Mayank Ph.D.
About ROCK MAKER®
ROCK MAKER is a powerful, yet easy-to-use software program that manages the entire protein crystallization process from experiment design including automated image scheduling, to liquid dispensing, image viewing and scoring, and data analysis. The software integrates seamlessly with our imagers and liquid handlers for a complete automated solution.
Over the past decade, the use of automation and specialized software for protein crystallization have become general practice in both academic and industrial settings alike. Advancements in both crystallization automation and software have allowed for higher experiment throughput and an increase in the number and rate of structures determined. With automated imaging systems, researchers can now run even more experiments and program a schedule for capturing images of their protein drops to see how those conditions are performing over time. One bottleneck to this process is manually scoring those protein drop images, which is often very time consuming and tedious.
The latest version of RM - v3.15 - brings MAchine Recognition of Crystallization Outcomes (MARCO), a well-known auto-scoring algorithm for visible images. MARCO is known to determine whether or not your images from the visible light path contain crystals with almost 94% accuracy1, saving you time and removing the guesswork from scoring images. Any images collected by ROCK IMAGER will automatically be scored by MARCO as a probability (between 0-1) of the presence of crystals in a particular image. MARCO was developed by the curious minds of a few Google employees in collaboration with a consortium of various labs from around the world, including GSK (USA), HWI (USA), CSIRO (Australia), and Bristol-Myers Squibb (USA).
1. Bruno AE, Charbonneau P, Newman J, Snell EH, So DR, Vanhoucke V, Watkins CJ, Williams S, Wilson J; Classification of crystallization outcomes using deep convolutional neural networks; PLoS One; 2018 Jun 20; 13(6): e0198883.