Deep clustering method facing single particle cryo-electron microscope images

Abstract

To solve the technical problem that the existing particle image classification methods have too much time overhead and low accuracy, the present invention provides a deep clustering method facing single particle cryo-electron microscope image. The deep clustering method comprises the following steps: a first step, performing data preprocessing, and sending data to an autoencoder to perform pre-training; a second step, training the autoencoder: clustering output vector features of the encoder; calculating a loss function by using the clustering result; and optimizing the autoencoder weight by using the stochastic gradient descent method; and a third step, inputting all the particle image data into the autoencoder, obtaining the clustering result and analyzing the clustering accuracy rate, determining whether the loss function and the accuracy rate change are less than a threshold, outputting the clustering result if so and performing ending, otherwise, returning to the second step. Theinvention can perform pre-training under various noise data to improve the noise reduction capability of a network, and adaptively train the weight of the loss function by using the stochastic gradient descent method to further improve the classification accuracy rate.

Description

Technical field [0001] The invention belongs to the field of machine learning, and particularly relates to a deep clustering method for single-particle cryo-electron microscope images. Background technique [0002] As the basic technology of high-resolution structural biology research, cryo-electron microscopy is the most popular structural biology research method in recent years. After years of development, cryo-electron microscopy technology has made breakthroughs in recent years, and it has become an effective method for studying the structure and function of biological macromolecules. The classification of two-dimensional particle images is an important step to obtain three-dimensional structure. Its main goal is to eliminate the errors of particle image rotation and translation, and classify the images according to the principles of compactness and discretization between classes, and finally particles belonging to the same class The image is averaged pixel by pixel. The re...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the technical problems of too much time consumption and low accuracy of the existing particle image classification technology, the present invention provides a deep clustering method for single particle cryo-EM images

Images