NEURO-FUZZY CONTROL FOR CRYO ELECTRON MICROSCOPY DATA ACQUISITION ON A JEOL TRANSMISSION ELECTRON MICROSCOPE

摘要

We report the development of software system involving a novel neuro-fuzzy control scheme that automates single-particle cryo electron microscopy (cryo-EM) data acquisition on modern JEOL Transmission Electron Microscopes (TEM). Data acquisition in TEM is the first crucial step during the single-particle analysis workflow. Importantly, the demand for large number of two dimensional images requires reliable and efficient automation of image data collection. The software system described here was developed to overcome the vulnerability of other software packages to non- ideal properties of TEM microscopes, such as specimen drift or non-reproducible sample holder movements. The making of this system involved two main steps: the development of an user-friendly graphical user interface (GUI) for the remote control of the electron microscope and the digital camera; and the use of an intelligent system for the sequential selection of hole areas in prefabricated EM grids. The software system uses different computer vision algorithms and provides useful information (regarding contrast transfer function analysis, ice-embedded single-particles detection, and ice quality validation) in real time. This work describes a different approach to partially automate single-particle cryo-EM data acquisition under low dose conditions, and also presents a neuro-fuzzy logic control system with neuro-adaptive learning capabilities. A Mamdani-type fuzzy inference system (FIS) takes decisions during the microscopy session as part of the control system by the use of a set of fuzzy IF-THEN control rules that replace microscope-expert's knowledge. The fuzzy inference system (FIS) was updated according to the given input/output data set obtained during different cryo-EM data acquisition sessions. Two adaptive network-based fuzzy inference systems (ANFIS) were trained to automate the selection of valid single-particles from each cryo-EM image and to detect ice-contamination in real-time. The novelty of this approach relies on the combination of computer vision with complex fuzzy logic algorithms, a combination that allows the system to choose the best strategy to overcome several different problems during a routine microscopy session. The application to cryo-EM data acquisition on our non-ideal electron microscope demonstrates the goodness and effectiveness of this intelligent software system to navigate through prefabricated EM grids as an experienced microscopist. The method is validated in real-time cryo-EM data acquisition for single particle approach of bacterial ribosomes.