Ng, Jia Tsing, Dekker, Carien, Kroemer, Markus, Osborne, Michael and von Delft, Frank (2014) Using Textons to Rank Crystallization Droplets by Likely Presence of Crystals. Acta Crystallographica Section D, D70 (10). pp. 2702-2718. ISSN 1399-0047

Abstract

Visual inspection of crystallization experiments is an important yet time-consuming and subjective step in X-ray crystallography. Previous published studies focused on automatically classifying crystallization droplets into distinct experiment outcomes; here, a method is described that instead ranks droplets by their likelihood of containing crystals or microcrystals, thereby prioritizing for visual inspection those images that are most likely to contain useful information. The use of textons is introduced to describe crystallization droplets objectively, allowing them to be scored with the posterior probability of a Random Forest classifier trained against droplets manually annotated for presence or absence of crystals or microcrystals. When images are ranked for viewing according to these scores, so that droplets with probable crystalline behaviour are placed early in the viewing order, then the top 10 wells include at least one human-annotated crystal or microcrystal for 94% of plates in a dataset of 196 plates imaged with a Minstrel HT system. This algorithm is robustly transferable to at least one other imaging system: applying the same parameters trained from Minstrel HT images to a dataset imaged by the Rock-Imager system, human-annotated crystals ranked in the top 10 wells for 90% of plates. Because the shape of the curve of scores is itself a useful overview of the plate’s behaviour, a custom viewer was written to integrate presentation of this curve with the display of images in their ranked order. Evidence is presented that such ranked viewing of images results in faster but more accurate evaluation of drops, in particular for the identification of microcrystals.

Item Type:	Article
Date Deposited:	13 Oct 2015 13:12
Last Modified:	04 Jul 2016 23:45
URI:	https://oak.novartis.com/id/eprint/22120