Expert Systems With Applications 55 (2016) 403–416 Contents lists available at ScienceDirect Expert Systems With Applications journal homepage: www.elsevier.com/locate/eswa Real-time and decision taking selection of single-particles during automated cryo-EM sessions based on neuro-fuzzy method David Gil-Carton a,∗ , Miguel Zamora a , James D. Sutherland c , Rosa Barrio c , Izaskun Garrido b , Mikel Valle a , Aitor J. Garrido b a Structural Biology Unit, Cooperative Center for Research in Biosciences CIC bioGUNE, Bizkaia Technology Park, 48160 Derio, Spain b Automatic Control Group – ACG, Department of Automatic Control and Systems Engineering, University of the Basque Country (UPV-EHU), 48013 Bilbao, Bizkaia, Spain c Functional Genomics Unit, Cooperative Center for Research in Biosciences CIC bioGUNE, Bizkaia Technology Park, 48160 Derio, Spain a r t i c l e i n f o Keyword: Cryo-electron microscopy Decision support systems Fuzzy logic Image processing Single-particle analysis a b s t r a c t Cryo-electron microscopy (cryo-EM) is a three-dimensional (3D) averaging technique that makes use of two-dimensional (2D) images of biological macromolecules preserved in a thin layer of vitreous ice. Re- cent advances in the field have facilitated the evolution of cryo-EM towards atomic resolution, and the technique provides 3D maps with detailed description of biological macromolecules. Data acquisition at the transmission electron microscope (TEM) is the first crucial step during the single-particle analysis workflow in cryo-EM. In order to exploit the potential of this structural technique for atomic or near- atomic resolution, the initial collection must allow recording of large datasets and, hence, requires op- erating the TEM in automated mode. The quality of the acquired dataset relies, however, on the exper- tise of researchers and unsupervised operations might result in low data quality. This work presents the first expert system integrated in a novel scheme to automate cryo-EM data acquisition in a TEM. This development takes advantage of fuzzy logic systems to integrate the working mode of an expert in a linguistic manner and to learn from acquired data through an adaptive network. A new method based on different image-processing algorithms and on adaptive neuro-fuzzy inference systems (ANFIS) iden- tifies, in an unsupervised manner, the single-particles present in cryo-EM images during the automated acquisition on a TEM. This single-particle identification system is integrated in a new intelligent control scheme to automate cryo-EM data acquisition. A classic fuzzy inference system (FIS) was programmed to make appropriate decisions during the session. The designed system can be trained for a specific sample and allows for unsupervised but efficient data collection imitating the working mode of an experienced microscopist. © 2016 Elsevier Ltd. All rights reserved. 1. Introduction In recent years cryo-EM single-particle analysis (Frank, 2006) has become a high-resolution technique (Kühlbrandt, 2014) due to the development of direct electron detectors and new proce- dures in image/video processing (Bai, McMullan, & Scheres, 2015; Bammes, Rochat, Jakana, Chen, & Chiu, 2012; Campbell et al., 2014; McMullan, Faruqi, Clare, & Henderson, 2014; Ruskin, Yu, & Grig- orieff, 2013). Consequently, there is a need for high-throughput and automation of modern microscopes to generate large datasets ∗ Corresponding author. Tel.: +34 946572516; fax: +34 944061301. E-mail addresses: dcarton@cicbiogune.es, davidgilcarton@yahoo.es (D. Gil- Carton), mzamora@cicbiogune.es (M. Zamora), jsutherland@cicbiogune.es (J.D. Sutherland), rbarrio@cicbiogune.es (R. Barrio), izaskun.garrido@ehu.es (I. Garrido), mvalle@cicbiogune.es (M. Valle), aitor.garrido@ehu.es (A.J. Garrido). of images to exploit this potential for near atomic resolution. Current available methods for automated acquisition in cryo-EM, such as the commercial EPU (E Pluribus Unum) software, or other academic software packages (Carragher et al., 2000; Zhang et al., 2009), allow collecting images from pre-selected areas. However, none of these softwares makes decisions during an automated session according to the quality of the cryo-EM images being recorded. Until now, the automatic evaluation of the data being collected during cryo-EM data acquisition sessions has not been available and, in the case of an unsupervised TEM session, the final output might be of poor quality. The presence of contamination, an inadequate ice thickness, empty holes due to broken carbon film, aggregation, poor concentration of particles, or damaged speci- mens, might severely reduce the final size of useful dataset. In this context, the development of new intelligent control schemes that imitates the behavior of expert researchers during cryo-EM data http://dx.doi.org/10.1016/j.eswa.2016.02.018 0957-4174/© 2016 Elsevier Ltd. All rights reserved.