Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
Review
. 2023 Aug 27;14(9):1674.
doi: 10.3390/mi14091674.

Novel Artificial Intelligence-Based Approaches for Ab Initio Structure Determination and Atomic Model Building for Cryo-Electron Microscopy

Megan C DiIorio  1 , Arkadiusz W Kulczyk  1   2
Affiliations
Review

Novel Artificial Intelligence-Based Approaches for Ab Initio Structure Determination and Atomic Model Building for Cryo-Electron Microscopy

Megan C DiIorio et al. Micromachines (Basel). .

Abstract

Single particle cryo-electron microscopy (cryo-EM) has emerged as the prevailing method for near-atomic structure determination, shedding light on the important molecular mechanisms of biological macromolecules. However, the inherent dynamics and structural variability of biological complexes coupled with the large number of experimental images generated by a cryo-EM experiment make data processing nontrivial. In particular, ab initio reconstruction and atomic model building remain major bottlenecks that demand substantial computational resources and manual intervention. Approaches utilizing recent innovations in artificial intelligence (AI) technology, particularly deep learning, have the potential to overcome the limitations that cannot be adequately addressed by traditional image processing approaches. Here, we review newly proposed AI-based methods for ab initio volume generation, heterogeneous 3D reconstruction, and atomic model building. We highlight the advancements made by the implementation of AI methods, as well as discuss remaining limitations and areas for future development.

Keywords: AI; AlphaFold2; artificial intelligence; cryo-EM; cryo-electron microscopy; deep learning; machine learning; neural networks.

PubMed Disclaimer

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Figure 1
Figure 1
A typical SPA workflow. Movies collected with an electron microscope are first motion-corrected. In this step, frames are aligned and averaged to account for beam-induced motion, which increases the SNR of images. The resultant micrographs undergo contrast transfer function (CTF) estimation to calculate the effects of defocus and microscope aberrations. This step is followed by particle picking and extraction, in which particles are selected and extracted from micrographs. The extracted particles are sorted based on orientation into discrete, 2D classes, and the user removes classes containing non-particles, noise, and artifacts. Such filtered particle stacks are used to generate one or more low-resolution ab initio reconstructions that are iteratively refined through 3D classification and 3D refinement to yield final Coulomb potential maps. Given sufficient map resolution and quality, atomic models can be built and validated.
Figure 2
Figure 2
Schematic of a basic ANN node. The node receives inputs (x1, x2, x3, …xn) from a preceding layer in the network (yellow). Each input is multiplied by a corresponding weight (w1, w2, w3, …wn, respectively). The node computes the sum of the weighted inputs and applies an activation function that generates the cumulative output of the node (y). This output is then transmitted to subsequent network layers for further processing.
Figure 3
Figure 3
Schematic of different DL architectures. (A) Convolutional Neural Network. In a CNN, a convolutional filter, or kernel, (red box) slides over the input data and extracts different features of the data, generating a corresponding feature map. Many maps are generated. These feature maps are then downsampled in the pooling layer to produce a “flattened” 1D image vector, which subsequently serves as the input to the fully connected layer where classification occurs. (B) Autoencoder. In the autoencoder architecture, the encoder NN (blue circles representing individual nodes in the network) transforms input data to a lower-dimensional, simplified latent representation (green circles). The decoder network (shown in purple) converts the latent representation back to the original dimension and form of the input. (C) Generative Adversarial Network. In a GAN, the generator aims to produce synthetic images that closely resemble the input data. The generator initiates this process with a latent variable (orange), which consists of a vector of random values. By adjusting the values of the latent variable, the generator can produce an array of synthetic outputs, thereby exploring different variations in the generated samples. The real, experimental images and the generated images are both provided as inputs to the discriminator network, which evaluates whether the images are real or not. During training, both networks update their weights based on the generator’s ability to produce realistic images and the discriminator’s ability to accurately decipher between the real and synthetic images.
Figure 4
Figure 4
In cases of high-resolution maps, atomic models can be built using tools originally designed for X-ray crystallographers. In an iterative process, the atomic model is refined using Phenix [128], followed by manual adjustments in Coot [130] to address issues such as interatomic clashes or geometry restraints. Here, a tyrosine sidechain rotamer is fit manually into the EM density using Coot.
Figure 5
Figure 5
The atomic model built from the cryo-EM map of a feline coronavirus spike protein [162] using DeepTracer [60]. (A) The 3.3 Å cryo-EM map of a feline coronavirus spike protein (EMDB ID: EMD-9891) that contains 1403 residues. (B) Density map fitted with the DeepTracer model. The DeepTracer model was built in just 14 min, compared to over 60 h required for model building with Phenix [60]. (C) Visualization of individual backbone atoms and side chains fitted to the cryo-EM Coulomb map using the molecular model obtained with DeepTracer.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. DiIorio M.C., Kulczyk A.W. Exploring the Structural Variability of Dynamic Biological Complexes by Single-Particle Cryo-Electron Microscopy. Micromachines. 2022;14:118. doi: 10.3390/mi14010118. - DOI - PMC - PubMed
    1. DiIorio M.C., Kulczyk A.W. A Robust Single-Particle Cryo-Electron Microscopy (cryo-EM) Processing Workflow with cryoSPARC, RELION, and Scipion. J. Vis. Exp. 2022;179:e63387. - PubMed
    1. Kulczyk A.W., Moeller A., Meyer P., Sliz P., Richardson C.C. Cryo-EM structure of the replisome reveals multiple interactions coordinating DNA synthesis. Proc. Natl. Acad. Sci. USA. 2017;114:E1848–E1856. doi: 10.1073/pnas.1701252114. - DOI - PMC - PubMed
    1. Kulczyk A.W., Sorzano C.O.S., Grela P., Tchorzewski M., Tumer N.E., Li X.P. Cryo-EM structure of Shiga toxin 2 in complex with the native ribosomal P-stalk reveals residues involved in the binding interaction. J. Biol. Chem. 2023;299:102795. doi: 10.1016/j.jbc.2022.102795. - DOI - PMC - PubMed
    1. Burley S.K., Berman H.M., Chiu W., Dai W., Flatt J.W., Hudson B.P., Kaelber J.T., Khare S.D., Kulczyk A.W., Lawson C.L., et al. Electron microscopy holdings of the Protein Data Bank: The impact of the resolution revolution, new validation tools, and implications for the future. Biophys. Rev. 2022;14:1281–1301. doi: 10.1007/s12551-022-01013-w. - DOI - PMC - PubMed

LinkOut - more resources