Deep convolutional neural networks for detecting secondary structures in protein density maps from cryo-electron microscopy

Li, Rongjian; Si, Dong; Zeng, Tao; Ji, Shuiwang; He, Jing

doi:10.1109/bibm.2016.7822490

Cited by 54 publications

(37 citation statements)

References 37 publications

(36 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This RMSD method walks each predicted backbone trace and pairs it with the closest Cα atom in the ground truth structure. This produces lower/better RMSD values than other methods [12] [25] because it allows for Cα skips in the ground truth backbone trace.…”

Section: Resultsmentioning

confidence: 99%

Cascaded-CNN: Deep Learning to Predict Protein Backbone Structure from High-Resolution Cryo-EM Density Maps

Moritz

Pfab

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

Cryo-electron microscopy (cryo-EM) has become a leading technology for determining protein structures. Recent advances in this field have allowed for atomic resolution. However, predicting the backbone trace of a protein has remained a challenge on all but the most pristine density maps (< 2.5Å resolution). Here we introduce a deep learning model that uses a set of cascaded convolutional neural networks (CNNs) to predict Cα atoms along a protein's backbone structure. The cascaded-CNN (C-CNN) is a novel deep learning architecture comprised of multiple CNNs, each predicting a specific aspect of a protein's structure. This model predicts secondary structure elements (SSEs), backbone structure, and Cα atoms, combining the results of each to produce a complete prediction map. The cascaded-CNN is a semantic segmentation image classifier and was trained using thousands of simulated density maps. This method is largely automatic and only requires a recommended threshold value for each evaluated protein. A specialized tabu-search path walking algorithm was used to produce an initial backbone trace with Cα placements. A helix-refinement algorithm made further improvements to the α-helix SSEs of the backbone trace. Finally, a novel quality assessment-based combinatorial algorithm was used to effectively map Cα traces to obtain full-atom protein structures. This method was tested on 50 experimental maps between 2.6Å and 4.4Å resolution. It outperformed several state-of-the-art prediction methods including RosettaES, MAINMAST, and a Phenix based method by producing the most complete prediction models, as measured by percentage of found Cα atoms. This method accurately predicted 88.5% (mean) of the Cα atoms within 3Å of a protein's backbone structure surpassing the 66.8% mark achieved by the leading alternate method (Phenix based fully automatic method) on the same set of density maps. The C-CNN also achieved an average RMSD of 1.23Å for all 50 experimental density maps which is similar to the Phenix based fully automatic method. This model and all code can be downloaded at https://github.com/DrDongSi/Ca-Backbone-Prediction. I. Introduction.Proteins perform a vast array of functions within organisms. From molecule transportation, to mechanical cellular support, to immune protection, proteins are the central building blocks of life in the universe [1]. Despite each protein being composed from a combination of the same 20 naturally occurring amino acids, a protein's functionality is mainly derived from its unique threedimensional (3D) shape. Therefore, learning the details of a protein's 3D structure is a prerequisite to understanding its biological function. A. Cryogenic Electronic Microscopy (Cryo-EM)Currently, one of the leading techniques for determining the atomic structure of proteins is cryoelectron microscopy (cryo-EM). Cryo-EM is a relatively new technique which uses a high-energy electron beam to image vitrified biological specimens. In the past five years, more than 1,000 protein structures have been determined to bet...

show abstract

Section: Resultsmentioning

confidence: 99%

Cascaded-CNN: Deep Learning to Predict Protein Backbone Structure from High-Resolution Cryo-EM Density Maps

Moritz

Pfab

et al. 2019

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…In literature, a number of machine learning techniques such as convolutional neural network (CNN) [34,35], deep convolutional neural network (DCNN) [36,37] recurrent neural network (RNN) [38], graphical model [39], have been proposed. Among these, CNN includes the fully connected layers which connect each neuron in a layer to all neurons in the next layer.…”

Section: Discussionmentioning

confidence: 99%

An Efficient Neural Network for Shape from Focus with Weight Passing Method

2018

View full text Add to dashboard Cite

In this paper, we suggest an efficient neural network model for shape from focus along with weight passing (WP) method. The neural network model is simplified by reducing the input data dimensions and eliminating the redundancies in the conventional model. It helps for decreasing computational complexity without compromising on accuracy. In order to increase the convergence rate and efficiency, WP method is suggested. It selects appropriate initial weights for the first pixel randomly from the neighborhood of the reference depth and it chooses the initial weights for the next pixel by passing the updated weights from the present pixel. WP method not only expedites the convergence rate, but also is effective in avoiding the local minimization problem. Moreover, this proposed method may also be applied to neural networks with diverse configurations for better depth maps. The proposed system is evaluated using image sequences of synthetic and real objects. Experimental results demonstrate that the proposed model is considerably efficient and is able to improve the convergence rate significantly while the accuracy is comparable with the existing systems.

show abstract

“…In addition to conventional machine learning techniques, very recently deep learning has been used for secondary structure prediction in EM maps. Deep learning, in particular 3-D CNN, turned out to be very suitable for identifying secondary structures from cryo-EM maps [104].…”

Section: Machine Learning Approachesmentioning

confidence: 99%

Advances in Structure Modeling Methods for Cryo-Electron Microscopy Maps

Alnabati

Kihara

2019

Molecules

View full text Add to dashboard Cite

Cryo-electron microscopy (cryo-EM) has now become a widely used technique for structure determination of macromolecular complexes. For modeling molecular structures from density maps of different resolutions, many algorithms have been developed. These algorithms can be categorized into rigid fitting, flexible fitting, and de novo modeling methods. It is also observed that machine learning (ML) techniques have been increasingly applied following the rapid progress of the ML field. Here, we review these different categories of macromolecule structure modeling methods and discuss their advances over time.Molecules 2020, 25, 82 2 of 13 methods, in this order. Then, we discuss methods that use machine learning approaches, which are emerging in recent years in the cryo-EM structure modeling field. Figure 1. The number of rigid fitting, flexible fitting, and de novo modeling software published per year. The statistics are based on publication. The plot shows 28 rigid fitting methods [9-36], 33 flexible fitting methods [37-69], and 8 de novo modeling methods [70-77]. Rigid Fitting MethodsIn rigid body fitting, high resolution atomic models which are derived from X-ray crystallography, Nuclear Magnetic Resonance (NMR) spectroscopy, or protein prediction are fitted into a cryo-EM map. One of the earliest rigid fitting methods is EMfit developed by Rossmann et al. in 2001 [12]. In EMfit, a high-resolution structure is placed manually into a specific position in the EM map. Then, a 3-D rotational search is applied to find the best orientation. After that, EMfit optimizes the initial fitting by performing local rotational and translational steps. In general, rigid-body fitting methods search for the best placement of an atomic model in a density map. Search algorithms that have been used for rigid fitting include Fast Fourier transform-based (FFT) [14,32,35], grid-threading Monte Carlo (GTMC) [16], spherical harmonic-based search [20], and geometric hashing [27]. FFT is a fast search scheme that accelerates the 3-D translational search [14]. HermiteFit speeds up the rotation step in the FFT by representing densities as three-dimensional orthogonal Hermite functions and performing rotation in the Hermite space [32]. Fast polar Fourier search is a variation of FFT, which is based on non-uniform SO(3) Fourier Transforms [35]. Its principal advantage is the ability to search efficiently and uniformly over a set of samples of the conformational space. GTMC combines grid search and Monte Carlo sampling [16]. GTMC divides the search space into grid points and uses Monte Carlo to find local maxima near the grid points to identify the global maximum. ADP_EM is a spherical harmonic-based (SH) method which applies exhaustive translational scanning and accelerates the rotational search by representing densities as SH functions [20]. Geometric hashing identifies a set of possible transformations which are stored in a fast-searchable hash map [27]. Later, the set of transformations is searched to find the best fit.While the methods above e...

show abstract

Deep convolutional neural networks for detecting secondary structures in protein density maps from cryo-electron microscopy

Cited by 54 publications

References 37 publications

Cascaded-CNN: Deep Learning to Predict Protein Backbone Structure from High-Resolution Cryo-EM Density Maps

Cascaded-CNN: Deep Learning to Predict Protein Backbone Structure from High-Resolution Cryo-EM Density Maps

An Efficient Neural Network for Shape from Focus with Weight Passing Method

Advances in Structure Modeling Methods for Cryo-Electron Microscopy Maps

Contact Info

Product

Resources

About