Device for estimating atomic structure, method for estimating atomic structure, and program for estimating atomic structure

Abstract

Provided is a device for estimating the atomic structure of a biomolecule, the device comprising: a generation unit that generates an image generated on the basis of a target biomolecule; an acquisition unit that acquires a plurality of first candidate structure images on the basis of two-dimensional projection images from a plurality of atomic structures prepared; a global search unit that, on the basis of the similarity between the generated image and the candidate structure images, selects, as a basic candidate structure, a structure having a high degree of structural similarity and / or molecular orientation similarity to the target biomolecule from among the plurality of atomic structure candidate structures; and a local search unit that acquires a plurality of second candidate structure images on the basis of two-dimensional projection images of a plurality of modified atomic structures obtained by modifying the basic candidate structure, and selects, on the basis of the similarity between the second candidate structure images and the generated image, a structure having a high degree of structural similarity and / or molecular orientation similarity to the target biomolecule from among the modified atomic structures.

Description

Atomic structure estimation device, atomic structure estimation method, and atomic structure estimation program

[0001] The present invention relates to an atomic structure estimation device, an atomic structure estimation method, and an atomic structure estimation program.

[0002] There is a need for methods to explore the atomic structures of large molecules, such as biomolecules. Biomolecules are highly flexible, and it has been difficult to elucidate their diverse structures and state transitions from observational images alone. Non-Patent Document 1 states, "Single-particle structural analysis using cryo-electron microscopy is attracting attention as a new method for determining protein structure. This method places few restrictions on the sample, and can be used to analyze everything from proteins of tens of kilodaltons to virus particles of hundreds of megadaltons (Figure 1). Furthermore, taking advantage of this spatial freedom, it is possible to analyze the dynamic structural changes and kinetics of proteins in response to various stimuli." (Abstract) Non-Patent Document 2 states, "Using cryoDRGN, we uncovered residual heterogeneity in high-resolution datasets of the 80S ribosome and the RAG complex, revealed a new structural state of the assembling 50S ribosome, and visualized large-scale continuous motions of a spliceosome complex. CryoDRGN contains interactive tools to visualize a dataset's distribution of per-particle variability, generate density maps for exploratory analysis, extract particle subsets for use with other tools, and generate trajectories to visualize molecular motions." (Abstract) Non-Patent Document 3 states, "Cryo-electron microscopy (cryo-EM) is a powerful technique for determining the structure of proteins and other macromolecular complexes at near-atomic resolution.In single particle cryo-EM, the central problem is to reconstruct the three-dimensional structure of a macromolecule from 104-7 noisy and randomly oriented two-dimensional projections. However, the imaged protein complexes may exhibit structural variability, which complicates reconstruction and is typically addressed using discrete clustering approaches that fail to capture the full range of protein dynamics. Here, we introduce a novel method for cryo-EM reconstruction that extends naturally to modeling continuous generative factors of structural heterogeneity. This method encodes structures in Fourier space using coordinate-based deep neural networks, and trains these networks from unlabeled 2D cryo-EM images by combining exact inference over image orientation with variational inference for structural heterogeneity. We demonstrate that the proposed method, termed cryoDRGN, can perform ab initio reconstruction of 3D protein complexes from simulated and real 2D cryo-EM image data.It is described in the abstract that "To our knowledge, cryoDRGN is the first neural network-based approach for cryo-EM reconstruction and the first end-to-end method for directly reconstructing continuous ensembles of protein structures from cryo-EM images." Non-Patent Document 4 states that "Modeling flexible macromolecules is one of the foremost challenges in single-particle cryogenic-electron microscopy (cryo-EM), with the potential to illuminate fundamental questions in structural biology. We introduce Three-Dimensional Flexible Refinement (3DFlex), a motion-based neural network model for continuous molecular heterogeneity for cryo-EM data. 3DFlex exploits knowledge that conformational variability of a protein is often the result of physical processes that transport density over space and tend to preserve local geometry. From two-dimensional image data, 3DFlex enables the determination of high-resolution 3D density, and provides an explicit model of a flexible protein's motion over its conformational landscape."Experimentally, for large molecular machines (tri-snRNP spliceosome complex, translocating ribosome) and small flexible proteins (TRPV1 ion channel, αVβ8 integrin, SARS-CoV-2 spike), 3DFlex learns nonrigid molecular motions while resolving details of moving secondary structure elements. [Prior art documents] [Non-patent documents] [Non-patent document 1] Song Ji-hong, Murata Kazuyoshi, "Dynamic structural analysis of proteins using cryo-electron microscopy," J. Comput. Chem. Jpn., Vol. 17, No. 1, pp. 38-45 (2018) [Non-patent document 2] Ellen D. Zhong, Tristan Bepler, Bonnie Berger, Joseph H. Davis, "CryoDRGN: reconstruction of heterogeneous cryo-EM structures using neural networks," Nature Methods volume 18, pages 176-185 (2021), https: / / doi.org / 10.1038 / s41592-020-01049-4 [Non-patent document 3] Ellen D. Zhong, Tristan Bepler, Joseph H.Davis, Bonnie Berger “Reconstructing continuous distributions of 3D protein structure from cryo-EM images”, https: / / arxiv.org / abs / 1909.05215, [Submitted on 11 Sep 2019 (v1), last revised 15 Feb 2020 (this version, v3)] [Non-patent Document 4] Ali Punjani, David J. Fleet “3DFlex: determining structure and motion of "Flexible proteins from cryo-EM" Nature Methods volume 20, pages860-870 (2023), https: / / www.nature.com / articles / s41592-023-01853-8. General disclosure

[0003] A first aspect of the present invention provides an atomic structure estimation device for a biomolecule. The atomic structure estimation device includes a generation unit, an acquisition unit, a global search unit, and a local search unit. The generation unit generates a generated image based on a target biomolecule. The acquisition unit acquires a plurality of first candidate structure images based on two-dimensional projection images from a plurality of prepared atomic structures. The global search unit selects, from the plurality of atomic structures, a structure and / or molecular orientation highly similar to that of the target biomolecule as a basic candidate structure based on the similarity between the generated image and the first candidate structure image. The local search unit acquires a plurality of two-dimensional projection images from a plurality of modified atomic structures obtained by modifying the basic candidate structure as second candidate structure images, and selects, from the modified atomic structures, a structure and / or molecular orientation highly similar to that of the target biomolecule based on the similarity between the second candidate structure image and the generated image.

[0004] In the above, the acquisition unit may acquire a plurality of first candidate structure images based on two-dimensional projection images projected from a plurality of directions surrounding the atomic structure.

[0005] In the above, the acquisition unit may acquire a first candidate structure image based on a two-dimensional projection image of an atomic structure registered in the molecular structure database.

[0006] In the above, the acquisition unit may acquire a first candidate structure image based on a two-dimensional projection image of an atomic structure predicted from the amino acid sequence.

[0007] In the above, the acquisition unit may acquire a first candidate structure image based on a two-dimensional projection image of an atomic structure reproduced in a molecular dynamics calculation.

[0008] In the above, the local search unit may increase the degree of similarity to the target biomolecule by repeatedly modifying the basic candidate structure multiple times.

[0009] In the above, the local search unit may simulate the structural transition of the selected modified atomic structure by cascade selection molecular dynamics (PaCS-MD).

[0010] In the above, the global search unit may match a plurality of first candidate structure images with each of a plurality of generated images, and select a basic candidate structure based on the similarity that is the matching result.

[0011] In the above, the global search unit may select a basic candidate structure based on the highest, lowest, average, and / or median similarity of multiple first candidate structure images matched to each of the multiple generated images.

[0012] In the above, the global search unit may calculate the similarity between the generated image and the first candidate structure image by parallel calculation.

[0013] In the above, the generator may generate a generated image based on a particle image of the target biomolecule. The generator may reduce noise in the particle image. The generated image may be a particle image with reduced noise.

[0014] In the above, the generation unit may reduce noise in the particle image by generating the image using machine learning.

[0015] In the above, the generating unit may apply a Fourier transform to the noise-reduced particle image to generate a target wavenumber space image that is an image after the Fourier transform as a generated image, and the acquiring unit may apply a Fourier transform to the two-dimensional projection image of the atomic structure to acquire a wavenumber space image that is an image after the Fourier transform as a first candidate structure image.

[0016] In the above, the generating unit may remove dust images from the latent space.

[0017] In the above, the structure estimation device may further include an output unit that outputs a three-dimensional structure model of the modified atomic structure selected by the local search unit.

[0018] In the above, the structure estimation device may further include an evaluation unit that evaluates and outputs interactions between domains of the modified atomic structure selected by the local search unit.

[0019] In the above, the structure estimation device may further include an analysis unit that analyzes structural features of the basic candidate structure and / or the modified atomic structure selected by the local search unit by principal component analysis. The output unit may output a result of the principal component analysis by plotting it on a plane or in space.

[0020] In the above, the global search unit may attribute the generated image to one of the basic candidate structures based on the similarity, and the output unit may further display the number of attributes of the generated image to each of the basic candidate structures on a plane or in space.

[0021] In the above, the output unit may display the attribution numbers as a heat map on a plane or in space.

[0022] In the above, the output unit may display the structural transition of the target biomolecule as a moving image by successively displaying generated images belonging to adjacent basic candidate structures on a plane or in space.

[0023] A second aspect of the present invention provides an atomic structure estimation method. The atomic structure estimation method includes a generation step, an acquisition step, a global search step, and a local search step. The generation step may generate a generated image based on a target biomolecule. The acquisition step may acquire a plurality of two-dimensional projection images as first candidate structure images for a plurality of prepared atomic structures. The global search step may select, as a basic candidate structure, one of the plurality of atomic structures whose structure and / or molecular orientation is highly similar to that of the target biomolecule based on the similarity between the generated image and the candidate structure image. The local search step may acquire a plurality of two-dimensional projection images as second candidate structure images from a plurality of modified atomic structures obtained by modifying the basic candidate structure, and select, from the modified atomic structures whose structure and / or molecular orientation is highly similar to that of the target biomolecule based on the similarity between the second candidate structure image and the generated image.

[0024] A third aspect of the present invention provides a program. The program is executed by a computer, causing the computer to function as a generation unit, an acquisition unit, a global search unit, and a local search unit. The generation unit generates a generated image based on a target biomolecule. The acquisition unit acquires a plurality of two-dimensional projection images of a plurality of prepared atomic structures as first candidate structure images. The global search unit selects, from the plurality of atomic structures, a structure and / or molecular orientation highly similar to that of the target biomolecule based on the similarity between the generated image and the first candidate structure image, as a basic candidate structure. The local search unit acquires a plurality of two-dimensional projection images of a plurality of modified atomic structures obtained by modifying the basic candidate structure as second candidate structure images, and selects, from the modified atomic structures, a structure and / or molecular orientation highly similar to that of the target biomolecule based on the similarity between the second candidate structure image and the generated image.

[0025] The above summary of the invention does not list all of the features of the present invention, and subcombinations of these features may also be inventions.

[0026] 1 shows a configuration of an atomic structure estimation device 10 according to this embodiment. 2 shows an atomic structure estimation flow by the atomic structure estimation device 10 according to this embodiment. 3 shows examples of particle images before and after applying VAE. 4 shows an example of acquisition of a two-dimensional projection image by the acquisition unit 120. 5 shows an example of acquisition of a two-dimensional projection image by the acquisition unit 120. 6 shows an example of selection of a basic candidate structure by the global search unit 130. 7 shows an example of a result of a global structure search. 8 shows an example of a result of a local structure search. 9 shows an example of a three-dimensional structure model of a selected atomic structure according to this embodiment. 10 shows an example of a heat map showing the number of attributes of generated images. 11 shows an example of a state transition of a basic candidate structure. 12 shows an example of a three-dimensional density map for each particle image set assigned to multiple representative structures. 13 shows an example of a video display by the output unit 150. 14 shows an example of a video display by the output unit 150. 15 shows an example of a structure space and a structure. 16 shows an example of a search result for Target1. 17 shows an example of a search result for a representative cluster. 18 shows an example of an estimated structure of a wild type. 19 shows an example of an estimated structure of a mutant type. 22 illustrates an example computer 2200 in which aspects of the present invention may be embodied, in whole or in part.

[0027] The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the scope of the invention as claimed. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0028] 1 shows the configuration of an atomic structure estimation apparatus 10 according to this embodiment. The atomic structure estimation apparatus 10 estimates the atomic structure of a biomolecule that is the subject of structural analysis (hereinafter also referred to as a "target biomolecule") based on a prepared two-dimensional projection image of the atomic structure and a particle image. The atomic structure estimation apparatus 10 includes a generation unit 110, an acquisition unit 120, a global search unit 130, a local search unit 140, an output unit 150, an evaluation unit 160, and an analysis unit 170.

[0029] The generator 110 generates a generated image based on the target biomolecule. For example, the generator 110 may generate a generated image based on a particle image of the target biomolecule. The generated image may be a particle image obtained by observing the target biomolecule with a cryo-electron microscope (CryoEM).

[0030] The generated image may be an image obtained by processing the particle image. For example, the generation unit 110 may reduce noise in the particle image and use the noise-reduced particle image as the generated image. The generated image may also be an image that has been subjected to a Fourier transform.

[0031] The acquisition unit 120 prepares atomic structures as virtual molecules in advance. For example, the acquisition unit 120 may acquire a structure recorded in a molecular structure database, a structure predicted from an amino acid sequence, and / or a structure reproduced by molecular dynamics, and use this as the atomic structure to be used as a seed. Alternatively or in addition to this, the acquisition unit 120 may further expand the atomic structure by simulating structural changes of the atomic structure using a molecular dynamics calculation or a structure sampling engine. The acquisition unit 120 acquires multiple two-dimensional projection images of the multiple atomic structures. The acquisition unit 120 acquires a first candidate structure image based on the two-dimensional projection images.

[0032] The global search unit 130 globally searches for the approximate atomic structure of the target biomolecule and sets the searched atomic structure as a basic candidate structure. For example, the global search unit 130 first acquires a generated image from the generation unit 110 and a first candidate structure image from the acquisition unit 120. Based on the similarity between the generated image and the first candidate structure image, the global search unit 130 selects, from the multiple prepared atomic structures, one that has a high degree of similarity in structure and / or molecular orientation to the target biomolecule as the basic candidate structure.

[0033] The global search unit 130 may be realized by parallel computing. For example, the global search unit 130 may be realized by a supercomputer having a large number of computing nodes.

[0034] The local search unit 140 performs a more accurate local atomic structure search based on the basic candidate structure. The local search unit 140 may modify the basic candidate structure to obtain multiple modified atomic structures. For example, the local search unit 140 may modify the basic candidate structure while maintaining the biomolecular structure to obtain multiple modified atomic structures. The local search unit 140 may continuously modify the basic candidate structure while maintaining the biomolecular structure to obtain multiple modified atomic structures. For example, the local search unit 140 performs a selective molecular dynamics calculation based on the basic candidate structure to obtain multiple modified atomic structures.

[0035] The local search unit 140 obtains a plurality of second candidate structure images based on the two-dimensional projection images of the plurality of modified atomic structures. The second candidate structure images may be images obtained by processing the two-dimensional projection images. For example, the local search unit 140 may perform a Fourier transform on the two-dimensional projection images to obtain the second candidate structure images.

[0036] The local search unit 140 selects, from the modified atomic structures, one that is highly similar to the target biomolecule in at least one of structure and molecular orientation based on the similarity between the second candidate structure image and the generated image. For example, the local search unit 140 may select, from the modified atomic structures, one that is highly similar to the target biomolecule in both structure and molecular orientation. The local search unit 140 may be implemented by parallel computing. For example, the local search unit 140 may be implemented by a supercomputer having a large number of computing nodes.

[0037] The output unit 150 outputs information about the basic candidate structure selected by the global search unit 130 and / or the modified atomic structure selected by the local search unit 140. For example, the output unit 150 outputs a three-dimensional structural model of the modified atomic structure.

[0038] The evaluation unit 160 evaluates the chemical and / or physical properties of the base candidate structure and / or the modified atomic structure. For example, the evaluation unit 160 evaluates and outputs interactions between fragments (residues) and / or domains of the selected atomic structure of the modified atomic structure.

[0039] The analysis unit 170 analyzes the structural features of the base candidate structure and / or the modified atomic structure, for example, by principal component analysis.

[0040] In this way, the atomic structure estimation device 10 according to this embodiment performs matching based on a projection image of an atomic structure reproduced by molecular dynamics and a particle image, and then performs a search based on the matching results. According to this embodiment, a comprehensive analysis is performed using a single noisy particle image (e.g., a Cryo-EM particle image) based on a target biomolecule, thereby realizing single-image analysis that directly estimates the diverse three-dimensional atomic structures of biomolecules.

[0041] According to this embodiment, compared to methods such as multi-image analysis, which averages multiple Cryo-EM particle images and removes noise to increase the resolution of common structural parts, it is possible to identify three-dimensional atomic structures with higher resolution without losing diverse structures due to averaging.

[0042] In a method for reproducing a three-dimensional structure based on an averaged particle image, the structure of the target biomolecule loses the characteristics of its flexible parts due to the averaging. On the other hand, the method of this embodiment links each generated image to the individual atomic structure reproduced by molecular simulation, so that structural information about the biomolecule is not lost due to the averaging, allowing for more precise estimation of the structure of the target biomolecule. In particular, this embodiment allows for more accurate identification of the structure of highly flexible parts of the biomolecule (e.g., parts that are prone to losing structural information due to averaging).

[0043] Furthermore, according to this embodiment, by combining a global search and a local search, it is possible to efficiently search for various structures of a target biomolecule using limited computational resources without sacrificing accuracy. Furthermore, according to this embodiment, it is possible to output the three-dimensional structure of the target biomolecule.

[0044] The atomic structure estimation apparatus 10 may be realized by one apparatus or by multiple apparatuses (computers) with different roles assigned to each apparatus. When realized by multiple apparatuses, the atomic structure estimation apparatus 10 may be a server / client system, in which case at least a part of the functions of the atomic structure estimation apparatus 10 may be realized by the server apparatus and the remaining functions may be realized by the client apparatus. The client apparatus may only have the function of transmitting input from a user to the server apparatus and displaying output from the server apparatus, in which case the substantial functions of the atomic structure estimation apparatus 10 are realized only by the server apparatus.

[0045] At least a part of the atomic structure estimation apparatus 10 may be realized by a computing device such as a supercomputer. The device (computer) used for at least a part of the atomic structure estimation apparatus 10 may be a computer such as a PC (personal computer), a tablet computer, a smartphone, a workstation, a server computer, a cloud server, or a general-purpose computer, or may be a computer system in which multiple computers are connected.

[0046] Alternatively, the device (computer) used in the atomic structure estimation apparatus 10 may be a dedicated computer designed for each application, or may be dedicated hardware realized by a dedicated circuit. Although not specifically described below, the atomic structure estimation apparatus 10 may be provided with a memory / hard disk or the like to store information necessary for processing as appropriate, and the atomic structure estimation apparatus 10 may transmit information between each processing module, such as the generation unit 110 and the acquisition unit 120.

[0047] 2 shows an atomic structure estimation flow by the atomic structure estimation apparatus 10 according to this embodiment. The atomic structure estimation apparatus 10 estimates the atomic structure of a target biomolecule by executing steps S100 to S1000. Some of steps S100 to S1000 may be omitted. Other operations may be performed in addition to steps S100 to S1000.

[0048] Steps S100 to S900 may be an example of single-image analysis that estimates a three-dimensional atomic structure from particle images. Step S1000 may be an example of multi-image analysis that analyzes various structural states based on the results of single-image analysis, and an example of a video that displays the movement of target biomolecules.

[0049] In S100, the generation unit 110 acquires a particle image. The generation unit 110 observes the target biomolecule using a particle analysis method using a transmission electron microscope and acquires an image of the target biomolecule. For example, the generation unit 110 acquires an image of the target biomolecule using a cryo-electron microscope (CryoEM).

[0050] The generation unit 110 acquires multiple particle images of the target biomolecule. As a result, the generation unit 110 obtains images of the target biomolecule captured from various orientations. The particle image may refer to a so-called single particle image, but the target biomolecule is not limited to a single particle. The target biomolecule may be a single particle, or a complex in which multiple single particle structures are assembled, coordinated, or bound together.

[0051] For example, the target biomolecule may be the spike protein alone (a trimer consisting of the same three regions) or a complex of the spike protein and an antibody (composed of the trimer plus the number of bound antibodies). Below, we will explain an example in which the wild-type and mutant (D614G) spike protein from SARS-CoV-2 are used as the target biomolecule.

[0052] The generating unit 110 may reject some of the particle images. For example, the generating unit 110 may perform multi-stage two-dimensional classification on the particle images and reject images with a low probability of biomolecules being present in the images based on the class average images. As an example, the generating unit 110 may reject particle images that are not classified into any class.

[0053] Next, in S200, the generator 110 generates a generated image from the particle image. For example, the generator 110 may reduce noise in the particle image. For example, the generator 110 may reduce noise in the particle image by image generation using machine learning such as deep learning.

[0054] As an example, the generation unit 110 may input the particle image to a variational autoencoder (VAE) and obtain the output image as the generated image. The VAE is a combination of an encoder and a decoder, and can compress the particle image to the dimension of the latent space by encoding it with the encoder. The compressed data can then be decoded with the decoder to restore the image to its original size.

[0055] In the manifold hypothesis, the various structures that biomolecules such as proteins can take in high-dimensional space can be viewed as a lower-dimensional manifold called the latent space. By encoding and decoding using VAE, it is possible to reduce noise while preserving the essential features of particle images.

[0056] Figure 3 shows examples of particle images before and after applying VAE. By applying VAE to the images in the left column of Figure 3, the images in the right column with reduced noise are obtained.

[0057] The generator 110 may apply a Fourier transform to the noise-reduced particle image to generate a target wavenumber space image, which is an image after the Fourier transform. For example, the generator 110 may input the particle image to a VAE and apply a Fourier transform to the image output. The generator 110 may output the target wavenumber space image as a generated image.

[0058] Alternatively, the generator 110 may input the particle image to the VAE and output the output image as the generated image without applying the Fourier transform. Alternatively, the generator 110 may output the particle image itself as the generated image without performing the Fourier transform and noise reduction.

[0059] The generation unit 110 may use all of the noise-reduced particle images as they are, or alternatively, may select and use only representative images. For example, the generation unit 110 may perform clustering analysis based on the image distribution in the latent space when the particle images are input to the VAE, and classify a large number of particle images into clusters. An existing method such as X-means may be used for the clustering analysis.

[0060] The generation unit 110 may select the central image of each cluster as a representative image of that cluster and use only the representative image as a particle image for subsequent processing, thereby reducing the amount of processing required for similarity calculation, which will be described later.

[0061] The generating unit 110 may perform a process of excluding some clusters in the latent space as groups of dust images. For example, the generating unit 110 may exclude clusters whose size occupies the latent space is equal to or greater than a threshold and / or whose representative images are unclear from subsequent processing targets. The generating unit 110 may automatically determine clusters to exclude, or may exclude clusters specified by a user.

[0062] In S300, the acquisition unit 120 acquires multiple two-dimensional projection images of the atomic structure of a virtual molecule. The acquisition unit 120 acquires multiple atomic structures of known biomolecules (hereinafter also referred to as "virtual molecules") from an existing molecular structure database. Examples of molecular structure databases include the Protein Data Bank (PDB), the Electron Microscopy Data Bank (EMDB), and / or the AlphaFold Protein Structure Database (PSD).

[0063] The acquisition unit 120 may use a structure predicted from the amino acid sequence. The acquisition unit 120 acquires a two-dimensional projection image of the atomic structure predicted from the amino acid sequence. The acquisition unit 120 may acquire two-dimensional projection images projected from a plurality of orientations surrounding the atomic structure. The plurality of orientations may be predetermined orientations, for example, orientations spaced at equal angular intervals.

[0064] The acquisition unit 120 may expand the structure by simulating the movement of the virtual molecule from which the atomic structure has been acquired using various simulation methods (e.g., molecular dynamics calculations or the OFLOD method). The acquisition unit 120 may acquire a two-dimensional projection image of the atomic structure reproduced in the molecular dynamics calculations. The acquisition unit 120 may acquire two-dimensional projection images projected from multiple orientations surrounding the atomic structure. The multiple orientations may be predetermined orientations, for example, orientations spaced at equal angular intervals.

[0065] The acquisition unit 120 may acquire multiple first candidate structure images based on multiple two-dimensional projection images. For example, if the generated image is a wavenumber space image that has been subjected to a Fourier transform, the acquisition unit 120 may apply a Fourier transform to the two-dimensional projection image of the atomic structure and acquire the wavenumber space image that is the image after the Fourier transform as the first candidate structure image. If the generated image has not been subjected to a Fourier transform, the acquisition unit 120 may use the two-dimensional projection image as the first candidate structure image as is.

[0066] Fig. 4 shows an example of acquisition of two-dimensional projection images by the acquisition unit 120. As shown in Fig. 4, the acquisition unit 120 may acquire images in which a virtual molecule 402 is projected two-dimensionally from a plurality of orientations (e.g., O01 to O06). Although six orientations are shown in Fig. 4 for the sake of convenience, the acquisition unit 120 may acquire two-dimensional projection images from an extremely large number of orientations (e.g., 10,000 orientations).

[0067] Fig. 5 shows an example of acquisition of two-dimensional projection images by the acquisition unit 120. As shown in Fig. 5, the acquisition unit 120 simulates the movement of a virtual molecule 402 over time using molecular dynamics calculations, and acquires two-dimensional projection images of the virtual molecule 402 at multiple times. For example, in Fig. 5, the acquisition unit 120 acquires two-dimensional projection images from the multiple directions shown in Fig. 4 at times T01, T02, T03, and so on.

[0068] Here, two-dimensional projection images acquired at the same time may be treated as belonging to the same atomic structure. However, the atomic structure changes over time, resulting in different structures at different times. For example, the atomic structure may be structure 1 at time T01, structure 2 at time T02, and structure 3 at time T03. Structures 1 to 3 are different structures.

[0069] For example, if 100 virtual molecules, 100 directions, and 100 times are set, the acquisition unit 120 acquires 100×100×100=1 million two-dimensional projection images.

[0070] When the target biomolecule is a complex, the virtual molecule can take on various states depending on the aggregation / coordination / binding state of each particle. Therefore, the acquisition unit 120 may prepare virtual molecules for each aggregation / coordination / binding state and perform simulations for each. For example, for a spike protein complex to which three antibodies are bound (i.e., a trimer + three antibodies), the acquisition unit 120 may prepare virtual molecules for a complex with no antibody bound at all, a complex with only one antibody bound at one site, a complex with antibodies bound at two sites, and a complex with antibodies bound at all three sites.

[0071] In S400 to S500, the global search unit 130 executes a global structure search that globally searches for the approximate atomic structure of the target biomolecule.

[0072] First, in S400, the global search unit 130 acquires the generated image obtained in S200 and the first candidate structure image obtained in S300. The global search unit 130 calculates the similarity between the generated image and the first candidate structure image. The global search unit 130 may calculate the similarity between the images using an existing template matching method such as cross-correlation, mean-square-error, squared-difference, or absolute-difference.

[0073] As an example, the global search unit 130 calculates c ｉｊ (ξ, α) may be calculated as the correlation value. According to Equations 1 to 3, the correlation value is calculated after normalizing the intensity image based on the wavenumber dependency of the expected value, so that correlations in a wider wavenumber range can be captured and even weaker signals can be used as effective signals. ...Equation 1 ...Equation 2 ...Equation 3

[0074] Here, the correlation value c ｉｊ An example of calculating the similarity using (ξ, α) will be described. ｃ，ｉｊ (k^ ｉｊ , α^ ｉｊ) may be calculated using Equations 4 to 7. The rotation of the molecule about the incident probe optical axis appears as a rotation about the center in the wavenumber space image.

[0075] By taking into account the rotational degree of freedom α, the correlation value when the pattern is rotated 360 degrees on a two-dimensional plane can be calculated at once using the correlation function. The correlation value displayed as a function of α and k = (2 / λ) sin(ξ / 2) is called a correlation diagram. Here, ξ = 2θ represents the scattering angle. Also, s is the wavenumber space intensity of the observed image, s is the intensity expectation value, λ is the wavelength of the incident probe light, and N ξ represents the discrete number of pixels on the concentric circles. ｉｊ The integral correlation value (or integral correlation diagram) I obtained by integrating (ξ, α) in the α and k directions ｃ，ｉｊ (kup, α) is defined. By performing this integration, the influence of noise can be reduced.

[0076] λ represents the wave number of the incident probe light. ｃ is the correlation length of the wavenumber spatial pattern and has a size equal to one molecular size. ｕｐ represents the maximum wave number of the wave number space image used in the similarity judgment calculation. In the integral correlation diagram, the direction of the correlation line is α^, and the maximum value of the integral correlation value when looking in the k direction at α^ is I ｃ，ｉｊ (k^ ｉｊ , α^ ｉｊ ) is obtained. By normalizing the wavenumber space images by the autocorrelation terms of the wavenumber space images obtained in the same manner, the similarity R ｃ，ｉｊ (k^ ｉｊ , α^ ｉｊ ) can be quantified.

[0077] ...Equation 4 ...Equation 5 ...Equation 6 ...Equation 7

[0078] The similarity between the target structure and the candidate structure is estimated through the wavenumber space image, but even if the candidate structure is the same, if the molecular orientation, which is the direction in which the probe light is incident, is different, the pattern of the wavenumber space image will be different. Therefore, to search for similar structures in wavenumber space, it is necessary to identify a molecular orientation similar to the target image (molecular orientation search). For each candidate structure, images with multiple different orientations are prepared, and the highest R obtained as a result of the molecular orientation search is calculated. ｃ The value can be taken as the structural identity of the candidate structure.

[0079] The global search unit 130 calculates the similarity between each of the multiple generated images and each of the first candidate structure images. For example, if there are N generated images and M first candidate structure images, the global search unit 130 may calculate the similarity between all or some of the combinations (N x M combinations) of the generated images and the first candidate structure images. The global search unit 130 may calculate the similarity between the generated images and the first candidate structure images by parallel calculation.

[0080] In S500, the global search unit 130 selects a base candidate structure from the plurality of virtual molecules. The global search unit 130 selects the base candidate structure based on the similarity calculated in S400. For each virtual molecule, the global search unit 130 may match a plurality of first candidate structure images with each of a plurality of generated images, and select the base candidate structure based on the similarity that is the matching result.

[0081] 6 shows an example of selection of a basic candidate structure by the global search unit 130. For example, the global search unit 130 calculates the similarity between a first candidate structure image 602 obtained by projecting a virtual molecule represented by Structure 1 from orientation O01 at time T01 and each of a plurality of generated images (generated image Img001, generated image Img002, generated image Img003, generated image Img004, generated image Img005, generated image Img006, etc.).

[0082] The global search unit 130 similarly calculates the similarity between the first candidate structure image 604, which is captured at the same time but in a different orientation from the first candidate structure image 602, and the multiple generated images. Because the time is the same, the first candidate structure image 602 and the first candidate structure image 604 are both treated as belonging to the same structure (structure 1).

[0083] At time TXX, which is the time after T01, the global search unit 130 also calculates the similarity between the first candidate structure image 606 and the multiple generated images. Here, since the structure has transitioned to structure XX with the passage of time, the first candidate structure image 606 is treated as belonging to structure XX.

[0084] The global search unit 130 comprehensively evaluates these similarities to calculate the structural compatibility, which is the degree to which the virtual molecule represented by each structure (e.g., "Structure 1") is similar to the target biomolecule in terms of structure and molecular orientation. For example, the global search unit 130 may determine the highest similarity (hereinafter also referred to as "total similarity of Structure 1") among all orientations (O01, O02, ...) for the virtual molecule represented by Structure 1 as the structural compatibility between Structure 1 and the target biomolecule.

[0085] For a virtual molecule represented by Structure 1, the global search unit 130 may determine the minimum value, average value, and / or median value of a predetermined number of the top similarities of Structure 1 as the structural compatibility between Structure 1 and the target biomolecule. The global search unit 130 may determine the number of generated images whose similarity exceeds a threshold as the structural compatibility between Structure 1 and the target biomolecule. For example, the global search unit 130 may determine the number of generated images whose similarity exceeds a threshold, or the total similarity of generated images whose similarity exceeds a threshold, as the structural compatibility between Structure 1 and the target biomolecule.

[0086] The global search unit 130 calculates the structural compatibility of the atomic structures of a plurality of virtual molecules (for example, structure 1, structure 2, structure 3, etc.) The global search unit 130 may select a basic candidate structure based on the structural compatibility.

[0087] For example, the global search unit 130 may select, as the basic candidate structure, a predetermined number of atomic structures having the highest structural compatibility and / or a structure having a structural compatibility exceeding a threshold value from among the plurality of atomic structures. In this manner, the global search unit 130 can select the basic candidate structure based on the highest, lowest, average, and / or median similarity of the plurality of first candidate structure images matched to each of the plurality of generated images.

[0088] As described above, the first candidate structure image and the generated image may be wavenumber space images. When the global search unit 130 calculates the similarity using a wavenumber space image, it is possible to evaluate the similarity of the structures by taking into consideration the translational and rotational degrees of freedom. If multiple basic candidate structures are selected, the processing from step S600 onward, which will be described later, may be executed in parallel for each basic candidate structure.

[0089] 7 shows an example of the results of the global structure search. The graph in the figure shows the results of a two-dimensional plot of the atomic structure features of multiple virtual molecules obtained by principal component analysis. Each gray point in the graph corresponds to the multiple virtual molecules prepared in S300. The global search unit 130 may select a basic candidate structure 702 from the multiple virtual molecules.

[0090] In S600 to S700, the local search unit 140 performs a local structure search of the target biomolecule based on the basic candidate structure. In the local structure search, the local search unit 140 increases the degree of similarity to the target biomolecule by repeatedly modifying the basic candidate structure. The local search unit 140 may continuously increase the degree of similarity to the target biomolecule. The local search unit 140 performs the local structure search while generating a new structure on-the-fly while maintaining the biomolecule structure.

[0091] If there are multiple basic candidate structures, the processes of S600 to S700 may be performed for each basic candidate structure. In this case, the processes may be performed in parallel for each basic candidate structure.

[0092] In S600, the local search unit 140 may perform a selection-type molecular dynamics calculation using a basic candidate structure as a starting structure. A plurality of modified atomic structures are obtained by modifying the basic candidate structure while maintaining the biomolecular structure based on the force field. A biomolecular structure is a structure that a biomolecule can take chemically, physically, and / or biologically, and may be a structure defined by pre-entered structural rules. For example, the local search unit 140 may modify the structure while maintaining the biomolecular structure by imposing restrictions on the bond terms (bond length, bond angle, dihedral angle, etc.) and non-bond terms (long-range electrostatic forces, van der Waals forces) of the atoms constituting the molecule. In a specific example, the local search unit 140 may use a Ramachandran plot to restrict the range of possible dihedral angles between atoms or groups of atoms in a biomolecular structure such as a protein.

[0093] The local search unit 140 may use a force field parameter set that physicochemically defines the likelihood of a biomolecular structure to be adopted, or a separately defined objective function. The local search unit 140 may perform modification while restricting the structures that the biomolecule can adopt, using the energy value or score value obtained from such an objective function as an index.

[0094] The modification of the base candidate structure is performed by a structural transition from the base candidate structure over time. As the selection-type molecular dynamics calculation, for example, a cascade selection-type molecular dynamics calculation (PaCS-MD) may be used. The local search unit 140 may simulate the structural transition of the modified atomic structure by the cascade selection-type molecular dynamics calculation (PaCS-MD).

[0095] A known algorithm may be used for PaCS-MD, for example, a method of repeating the following (A) and (B): (A) performing molecular dynamics (MD) calculations at a constant temperature from multiple modified atomic structures, and (B) ranking the structures based on their similarity to the target structure (similarity calculated in S700, described later), and using the top multiple structures (a predetermined number of structures) as new initial structures, restarting multiple short-term molecular dynamics calculations.

[0096] The local search unit 140 acquires a plurality of two-dimensional projection images of the simulated modified atomic structures. The local search unit 140 may acquire a plurality of two-dimensional projection images of the modified atomic structures in a manner similar to that of the acquisition unit 120 of S300. For example, the local search unit 140 may acquire two-dimensional projection images of each modified atomic structure from a plurality of orientations at a plurality of time points.

[0097] The local search unit 140 may acquire multiple second candidate structure images based on the two-dimensional projection image of the modified atomic structure. For example, if the generated image is a wavenumber space image that has been subjected to a Fourier transform, the local search unit 140 may apply a Fourier transform to the two-dimensional projection image of the modified atomic structure and acquire the wavenumber space image that is the image after the Fourier transform as the second candidate structure image. If the generated image has not been subjected to a Fourier transform, the local search unit 140 may use the two-dimensional projection image as the second candidate structure image as is.

[0098] In S700, the local search section 140 calculates the similarity between the second candidate structure image and the generated image obtained in S200. The local search section 140 may calculate the similarity using the same method as the global search section 130 in S400.

[0099] The local search unit 140 may calculate the similarity between the multiple second candidate structure images and the generated image for each modified atomic structure, and calculate the structural compatibility between the modified atomic structure and the target biomolecule based on the similarity. The local search unit 140 may calculate the structural compatibility from the similarity using a method similar to that used by the global search unit 130 in S500.

[0100] In S800, the local search unit 140 determines whether to terminate the local structure search. If the local search unit 140 determines that the search termination condition is met, it may terminate the local structure search for the basic candidate structure currently being searched. The search termination condition may be "the structural fitness of the modified atomic structure exceeds a threshold," "the degree of improvement in the structural fitness of the modified atomic structure in the repetition of S600 to S800 falls below a threshold," "S600 to S800 are repeated a predetermined number of times," and / or "a predetermined time has elapsed," etc.

[0101] The local search unit 140 may end the repetitions after an arbitrary number of repetitions input by the user.

[0102] When the local search unit 140 completes the local structure search for all basic candidate structures, it selects, from among the modified atomic structures searched up to that point, the one with the highest structural compatibility with the target biomolecule as the modified atomic structure. The modified atomic structure selected by the local search unit 140 is hereinafter also referred to as the "selected atomic structure." The local search unit 140 may select multiple modified atomic structures (e.g., a predetermined number of structures with the highest structural compatibility) as the selected atomic structures.

[0103] If the local search unit 140 terminates the local structure search, it advances the process to S900. Otherwise, the local search unit 140 returns the process to S600 and continues the local structure search. In the next step S600, the local search unit 140 may further modify the modified atomic structure with the highest structural fitness to create a new modified atomic structure and continue the search. In this way, the local search unit 140 may continuously improve the structural fitness by repeatedly modifying the atomic structure starting from the basic candidate structure.

[0104] Fig. 8 shows an example of the results of the local structure search. The graph in the figure is the result of plotting the characteristics of the atomic structure two-dimensionally using principal component analysis, as in Fig. 7. The black circles in the figure represent modified atomic structures generated by repeated modifications of the basic candidate structure 702. The local search unit 140 ultimately selects a selected atomic structure 802.

[0105] 8 shows that the local structure search starts from a base candidate structure 702 and gradually modifies the atomic structure. In this way, according to this embodiment, a general structure close to the target biomolecule is searched for by a global structure search, and then a local structure search is performed to search for a structure that is more suitable for the target biomolecule.

[0106] In the global and / or local search, instead of searching all directions (10,240) for each candidate structure, a small number of favorable molecular orientations may be searched for using Bayesian optimization, thereby reducing computational costs.

[0107] In S900, the output unit 150 outputs the results obtained by the search. For example, the output unit 150 outputs information about the base candidate structure, the modified atomic structure, and / or the selected atomic structure. As an example, the output unit 150 outputs a three-dimensional structural model of the selected atomic structure.

[0108] 9 shows an example of a three-dimensional structure model of a selected atomic structure according to this embodiment. The output unit 150 may output and display a three-dimensional structure model 902 of the selected atomic structure on an output screen. The output unit 150 may accept an operation on the three-dimensional structure model 902 from a user.

[0109] For example, the output unit 150 may input an operation such as rotation, enlargement, reduction, horizontal movement, approach, or separation of the three-dimensional structural model 902, and change the display of the three-dimensional structural model 902 in accordance with the operation. As an example, in response to a 90-degree rotation operation input by the user, the output unit 150 may output a three-dimensional structural model 904 obtained by rotating the three-dimensional structural model 902 by 90 degrees onto the output screen.

[0110] The output unit 150 may output the results of the analysis of the basic candidate structure, the modified atomic structure, and / or the selected atomic structure by the analysis unit 170. For example, the analysis unit 170 may analyze the structural features of the basic candidate structure, the modified atomic structure, and / or the selected atomic structure by principal component analysis. The output unit 150 may plot the results of the principal component analysis on a plane as shown in FIGS. 7 and 8 and output them. The output unit 150 may also plot the results of the principal component analysis in a three-dimensional or higher space and output them. By executing S100 to S900 in this manner, single image analysis is performed.

[0111] The output unit 150 may output the results of the evaluation of the interactions between the base candidate structure, the modified atomic structure, and / or the selected atomic structure by the evaluation unit 160. For example, the evaluation unit 160 may perform electronic state calculations using the fragment molecular orbital method (FMO) on the base candidate structure, the modified atomic structure, and / or the selected atomic structure obtained by the single image analysis.

[0112] In FMO, a molecule is divided into subunits called fragments, and the electronic state of the molecule is calculated using the electronic energy obtained from calculations on each fragment. For FMO calculations, known software such as ABINIT-MP [6-7] Open Ver. 1 Rev. 23Q2 can be used. Analysis of inter-fragment interactions can be performed using, for example, PIEDA (pair-interaction energy decomposition analysis).

[0113] The evaluation unit 160 may calculate the interactions between fragments (residues) and / or between domains of the base candidate structure, the modified atomic structure, and / or the selected atomic structure as an inter-domain IFIE map. For example, if the target biomolecule has six domains, the analysis unit 170 may generate a 6x6 tile heat map.

[0114] The evaluation unit 160 may calculate the interaction between a first domain and a second domain by analyzing the interaction between a fragment included in a first domain and a fragment included in another second domain. For example, the evaluation unit 160 may identify a pair of fragments in which the absolute value of any of the three components other than ES among the inter-fragment interaction energies (e.g., electrostatic interaction ES, exchange-repulsion interaction EX, charge-transfer interaction CT+mix, dispersion force interaction DI, etc.) in the fragment molecular orbital (FMO) method is equal to or greater than a predetermined value (e.g., 3 kcal / mol). For example, the evaluation unit 160 may calculate the inter-domain interaction by averaging the interactions of the identified pair of fragments for each domain.

[0115] 10 shows an example of state transitions of a base candidate structure, a modified atomic structure, and a selected atomic structure. The base candidate structure, the modified atomic structure, and the selected atomic structure are mainly classified into four states: a free state 1102, an open state 1104, a closed state 1106, and a condensed state 1108. Therefore, the target biomolecule is considered to be transitioning between these states. By performing an FMO calculation, the evaluation unit 160 can evaluate the degree to which interactions between fragments (residues) and / or domains of the selected atomic structure contribute to the stabilization of these states.

[0116] In S1000, the output unit 150 may perform multi-image analysis. In S400 or S500, the global search unit 130 may assign each of the generated images to any of all or some of the representative basic candidate structures based on the degree of similarity. For example, the global search unit 130 may assign each of the generated images to one or more basic candidate structures with high similarity. In S700 or S800, the local structure search may assign each of the generated images to any of all or some of the representative modified atomic structures and / or selected atomic structures. Here, the output unit 150 may display the number of generated images assigned to each of the basic candidate structures, modified atomic structures, and / or selected atomic structures on a plane or in space.

[0117] For example, the output unit 150 may display the number of attributions on a plane or space on which the results of the principal component analysis are plotted. For example, the output unit 150 may display the number of attributions as a heat map on a plane or space on which the results of the principal component analysis are plotted. This allows the atomic structure estimation device 10 to not only determine what structures the target biomolecule can take, but also determine in which state it will be stable (in which the number of attributions will be high).

[0118] 11A shows an example of a heat map showing the number of attributes of generated images. The output unit 150 may output a heat map such as that shown in FIG. 11A. Each point in FIG. 11A corresponds to a basic candidate structure selected as a representative. Each point is displayed with a different density depending on the number of attributes of generated images for the corresponding basic candidate structure. For example, the higher the number of attributes, the darker the point is displayed.

[0119] As shown in Figure 11A, generated images derived from one target biomolecule are assigned to multiple different representative structures based on similarity, indicating that the basic candidate structure can take on various states and transition between them.

[0120] A 3D density map can be constructed by performing 3D reconstruction using a set of generated images assigned to a representative structure or a set of original particle images associated with the generated images. As shown in Figure 11B, multiple diverse 3D density maps, such as those shown in M0-M17, can be constructed by performing 3D reconstruction for each set of particle images assigned to multiple representative structures. By performing a global search, more diverse 3D density maps can be constructed compared to the existing method, cryoDRGN.

[0121] The output unit 150 may output the movement of the target biomolecule as a video. For example, the output unit 150 may represent the movement of the target biomolecule as a video by continuously displaying three-dimensional structural models of multiple similar atomic structures or generated images attributable to them. As an example, the output unit 150 may display the structural transition of the target biomolecule as a video by continuously displaying adjacent basic candidate structures, modified atomic structures, and / or selected atomic structures (or generated images attributable to them) on a plane or space plotted by principal component analysis or the like.

[0122] 12 shows an example of a moving image displayed by the output unit 150. For example, the output unit 150 acquires a generated image 1222 belonging to atomic structure 33, a generated image 1224 belonging to atomic structure 17, and a generated image 1226 belonging to atomic structure 23, which are adjacent on the plane plotted by the principal component analysis.

[0123] 13 , the output unit 150 sequentially displays the generated image 1222, the generated image 1224, and the generated image 1226 in accordance with the positional relationship on the trajectory 1210. This allows the output unit 150 to generate a pseudo-movie of the target biomolecule whose state gradually changes.

[0124] At least a part of the flow of S100 to S1000 may be realized by parallel computing. For example, S200 and S400 to S700 may be executed in parallel by a supercomputer having a large number of computing nodes.

[0125] [Example] [Complex] The target biomolecule was a Fab complex, in which an infection-enhancing antibody (Fab) binds to the spike protein of SARS-CoV-2, and the Fab complex was observed and photographed using a transmission electron microscope. The spike protein has three Fab binding sites. Therefore, the observed particle images may contain a mixture of 0, 1, 2, and 3 Fabs. Based on the photographic data, a set of cryo-EM particle images (approximately 110,000 images, J326) of the SARS-CoV-2 spike protein Fab complex was obtained using the dedicated software RELION.

[0126] VAE (latent space dimension 12, Dropout = 0.3, noise = 0.3) was applied to the cryo-EM particle image set to obtain generated images with improved image quality. Furthermore, clustering analysis using the X-means method was performed on each of the improved images in the VAE latent space to obtain a representative image for each cluster. Fifteen clusters were obtained, but two clusters had unclear representative images and were widely distributed in the latent space. Representative images for 13 clusters were selected as target images for further processing.

[0127] In parallel, virtual molecules were prepared by MD. CryoEM images of Fab complexes can be a mixture of no Fab bound (referred to as "Fab0"), one Fab bound (referred to as "Fab1"), two Fab bound (referred to as "Fab2"), and three Fab bound (referred to as "Fab3"). For this purpose, four structural space patterns were prepared in advance using all-atom OFLOOD (AA-OFLOOD). For each binding state, AA-OFLOOD was run for 20 cycles with 250 replicas, and structural sampling was performed with a maximum of 500,000 structures preserved in each Fab state. Principal component analysis was performed on the 500,000 structures, and approximately 5,000 representative structures were selected on the PC1-PC2 plane.

[0128] As shown in Figures 14A and 14B (hereinafter collectively referred to as Figure 14), a diverse structural space with a maximum RMSD of 15 Å was constructed based on the initial structure. In Figure 14, the structural space and structural examples of Fab0 to Fab3 are plotted, and the structural distribution is projected onto the principal component axes (PC1-PC2), with the RMSD values ​​from the initial structure expressed as shading.

[0129] Similarity scores were calculated using the methods described in S400 to S500, etc., and basic candidate structures were selected. Specifically, for each 3D atomic structure generated by MD, similarity scores were comprehensively calculated between a group of 2D projection images from all directions (180 degrees divided into 10, 240 parts, taking into account symmetry in Fourier space) and experimental 2D images with improved image quality, and the 3D atomic model showing the highest similarity score was selected as the promising structure (i.e., basic candidate structure).

[0130] A local structure search was performed using the methods described in S600 to S800. Specifically, the promising structure obtained in the global search was used as the initial structure, and a local search was performed using 100 cycles of PaCS (Parallel Cascade Selection)-MD to find structures in its vicinity that more closely matched the target image, thereby obtaining the final model structure. In this local search, a modified PaCS-MD algorithm was applied, which used the similarity score in the FT image as a constraint, to efficiently search for a three-dimensional structure that showed a higher similarity score.

[0131] Figures 15A-I (hereinafter collectively referred to as Figure 15) show the search results obtained for Fab3, in which three antibodies bind to the spike protein. As shown in Figures 15G-H, a plausible three-dimensional structure was identified, showing a similarity of approximately 0.7. Target 1 in Figure 15 corresponds to one of the 15 clusters (Target 0 to Target 14).

[0132] As a result of performing a global structure search on the 13 noise-reduced target images, six clusters (Target 0, 1, 3, 4, 11, and 13) were found to be highly likely to correspond to Fab3, three clusters (Target 2, 6, and 9) were found to be highly likely to correspond to Fab2, four clusters (Target 5, 7, and 10) were found to be highly likely to correspond to Fab1, and no clusters were found to be highly likely to correspond to Fab0. Figure 16A shows the estimated structural positions in the structural space (PC1-PC2 plane) of each cluster corresponding to Fab3, as well as the estimated three-dimensional structures.

[0133] The electronic state was evaluated using the method explained in S1000 etc. For structures with high similarity, electronic state calculations were performed using the fragment molecular orbital method (FMO).

[0134] Figures 16B-G show the results of local structure search for representative clusters (Target 0, 1, 3, 4) along with domain IFIE maps obtained by FMO calculations. As shown in the figure, a close state (Target 1 shown in r) and a condensed state (Target 3 shown in s) were observed. A complex-specific pattern (Target 0, 4) that was not observed with the spike protein alone, as described below, was also observed. In the interaction with the Fab region, Fab ａ Not only Fab ｂ It was also observed that the above region simultaneously interacted with the NTD (Targtet1, 4).

[0135] [Spike protein alone] The wild-type and mutant (D614G) SARS-CoV-2 spike protein were used as target biomolecules and observed and photographed using a transmission electron microscope. Based on the photographic data, a set of cryo-EM particle images of the wild-type (approximately 190,000 images) and the mutant (approximately 120,000 images) were obtained using the dedicated software RELION.

[0136] VAE (latent space dimension 12, Dropout = 0.3, noise = 0) was applied to the cryo-EM particle image set to obtain generated images with improved image quality. Furthermore, clustering analysis using the k-means method was performed on each of the improved images, resulting in 38 wild-type and 36 mutant-type clusters. Based on the cluster representative image and the characteristics of each cluster distribution in the latent space (widely distributed, narrowly distributed), 26 wild-type and 26 mutant-type target images were selected.

[0137] Single-image analysis was performed on 26 target images of the wild-type protein, and the top 10 structures estimated from each target image (260 structures in total) were subjected to FMO calculations, IFIE, and PIEDA analysis. Four distinct interaction states were identified: the open state, the closed state, the condensed state, and the free state.

[0138] Figure 17A shows the estimated structural position in structural space (PC1-PC2 plane) for the wild-type protein, as well as the target image and estimated 3D structure. Figures 17B-E show the results of local search for four representative wild-type cases (Target 33, 5, 0, and 25) and domain IFIE maps obtained by FMO calculations. The inter-domain IFIE map is composed of 6 x 6 tiles, and the darker the color of each tile, the more stable the interactions between domains.

[0139] Single-image analysis was also applied to the 26 mutant target image groups, and FMO calculations, IFIE, and PIEDA analyses were performed on the top 10 structures estimated from each target image (a total of 260 structures). The success rate of FMO calculations was 78.8% (205 successful cases out of 260). The free state accounted for a high proportion of 46%, suggesting that a shift in the structural state occurs with the mutation. It was found that the D614G mutation stabilizes the RBD-raised structural state (open), tending to enhance viral infection.

[0140] Figure 18A shows the estimated structure positions in the structural space (PC1-PC2 plane) for the mutant, the target image, and the estimated 3D structure. Figures 18B to 18E show the results of local search for four representative cases of the mutant (Target 20, 22, 30, and 1) and domain IFIE maps obtained by FMO calculations.

[0141] As described above, according to this embodiment, by realizing single-image analysis, the atomic structures of multiple structural states of a target biomolecule can be identified with high accuracy using fewer computational resources based on experimental data through analysis in image space and structural space, and this can be output as a three-dimensional structure. According to this embodiment, it is possible to elucidate various structures of a target biomolecule. According to this embodiment, it is possible to elucidate and output state transitions between structures, interactions between fragments, etc. According to this embodiment, it is possible to elucidate various structural states of a biomolecule with high resolution based on experimental data. In addition / alternatively, it is possible to evaluate and / or output the elucidate various structural states using a unique method.

[0142] Various embodiments of the present invention may be described with reference to flowcharts and block diagrams, where the blocks may represent (1) stages of a process in which operations are performed or (2) sections of apparatus responsible for performing the operations. Particular stages and sections may be implemented by dedicated circuitry, programmable circuitry provided with computer-readable instructions stored on a computer-readable medium, and / or a processor provided with computer-readable instructions stored on a computer-readable medium. Dedicated circuitry may include digital and / or analog hardware circuitry, and may include integrated circuits (ICs) and / or discrete circuits. Programmable circuitry may include reconfigurable hardware circuitry including logical AND, OR, XOR, NAND, NOR, and other logic operations, flip-flops, registers, memory elements such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), and the like.

[0143] A computer-readable medium may include any tangible device capable of storing instructions that are executed by an appropriate device, such that the computer-readable medium having instructions stored thereon comprises an article of manufacture containing instructions that can be executed to create means for performing the operations specified in the flowcharts or block diagrams. Examples of computer-readable media may include electronic, magnetic, optical, electromagnetic, and semiconductor storage media. More specific examples of computer-readable media may include floppy disks, diskettes, hard disks, random access memories (RAMs), read-only memories (ROMs), erasable programmable read-only memories (EPROMs or flash memories), electrically erasable programmable read-only memories (EEPROMs), static random access memories (SRAMs), compact disc read-only memories (CD-ROMs), digital versatile discs (DVDs), Blu-ray discs, memory sticks, integrated circuit cards, and the like.

[0144] The computer readable instructions may include either assembler instructions, Instruction Set Architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, JAVA®, C++, etc., and conventional procedural programming languages ​​such as the “C” programming language or similar programming languages.

[0145] The computer-readable instructions may be provided to a processor or programmable circuitry of a programmable data processing apparatus, such as a computer, locally or over a local area network (LAN), a wide area network (WAN) such as the Internet, etc., which executes the computer-readable instructions to create means for performing the operations specified in the flowcharts or block diagrams.

[0146] Here, the computer may be a computer such as a PC (personal computer), a tablet computer, a smartphone, a workstation, a server computer, a cloud server, or a general-purpose computer, or may be a computer system in which multiple computers are connected. Such a computer system in which multiple computers are connected is also called a distributed computing system, which is a computer in the broad sense. In a distributed computing system, the multiple computers collectively execute a program by each executing a part of the program and transferring data between the computers as needed during program execution.

[0147] Examples of processors include computer processors, central processing units (CPUs), processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc. A computer may have one processor or multiple processors. In a multiprocessor system with multiple processors, each processor executes a portion of a program and passes data between processors as needed during program execution, allowing the multiple processors to collectively execute the program. For example, in multitasking, each of the multiple processors may execute a portion of each task in small chunks by switching tasks at each time slice. In this case, which portion of a program each processor executes changes dynamically. Alternatively, which portion of a program each of the multiple processors executes may be statically determined by multiprocessor-aware programming.

[0148] 19 illustrates an example of a computer 2200 in which aspects of the present invention may be embodied, in whole or in part. Programs installed on the computer 2200 may cause the computer 2200 to function as or perform operations associated with an apparatus or one or more sections of the apparatus according to embodiments of the present invention, and / or to perform a process or steps of a process according to embodiments of the present invention. Such programs may be executed by the CPU 2212 to cause the computer 2200 to perform specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0149] A computer 2200 according to this embodiment includes a CPU 2212, a RAM 2214, a graphics controller 2216, and a display device 2218, which are interconnected by a host controller 2210. The computer 2200 also includes input / output units such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card drive, which are connected to the host controller 2210 via an input / output controller 2220. The computer also includes legacy input / output units such as a ROM 2230 and a keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

[0150] The CPU 2212 operates according to programs stored in the ROM 2230 and RAM 2214, thereby controlling each unit. The graphics controller 2216 acquires image data generated by the CPU 2212 into a frame buffer or the like provided in the RAM 2214 or into the graphics controller 2216 itself, and causes the image data to be displayed on the display device 2218.

[0151] The communication interface 2222 communicates with other electronic devices via a network. The hard disk drive 2224 stores programs and data used by the CPU 2212 in the computer 2200. The DVD-ROM drive 2226 reads programs or data from the DVD-ROM 2201 and provides the programs or data to the hard disk drive 2224 via the RAM 2214. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0152] ROM 2230 stores therein a boot program or the like that is executed by computer 2200 upon activation, and / or programs that depend on the hardware of computer 2200. I / O chip 2240 may also connect various I / O units to I / O controller 2220 via parallel ports, serial ports, keyboard ports, mouse ports, etc.

[0153] The programs are provided by a computer-readable medium such as a DVD-ROM 2201 or an IC card. The programs are read from the computer-readable medium, installed in the hard disk drive 2224, RAM 2214, or ROM 2230, which are also examples of computer-readable media, and executed by the CPU 2212. Information processing described in these programs is read by the computer 2200, and brings about cooperation between the programs and the various types of hardware resources described above. An apparatus or method may be configured by implementing information manipulation or processing in accordance with the use of the computer 2200.

[0154] For example, when communication is performed between computer 2200 and an external device, CPU 2212 may execute a communication program loaded in RAM 2214 and instruct communication interface 2222 to perform communication processing based on the processing described in the communication program. Under the control of CPU 2212, communication interface 2222 reads transmission data stored in a transmission buffer processing area provided in RAM 2214, hard disk drive 2224, DVD-ROM 2201, or a recording medium such as an IC card, and transmits the read transmission data to the network, or writes received data received from the network to a reception buffer processing area or the like provided on the recording medium.

[0155] Furthermore, the CPU 2212 may cause all or a necessary portion of a file or database stored on an external recording medium such as the hard disk drive 2224, the DVD-ROM drive 2226 (DVD-ROM 2201), an IC card, etc. to be read into the RAM 2214, and may perform various types of processing on the data on the RAM 2214. The CPU 2212 then writes back the processed data to the external recording medium.

[0156] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and may undergo information processing. The CPU 2212 may perform various types of processing on data read from the RAM 2214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc., as described throughout this disclosure and specified by the instruction sequences of the programs, and write the results back to the RAM 2214. The CPU 2212 may also search for information in a file, database, etc. on the recording medium. For example, if multiple entries each having an attribute value of a first attribute associated with an attribute value of a second attribute are stored on the recording medium, the CPU 2212 may search for an entry that matches a condition specified by the attribute value of the first attribute from among the multiple entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0157] The above-described programs or software modules may be stored in a computer-readable medium on or near the computer 2200. A recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can also be used as a computer-readable medium, thereby providing the programs to the computer 2200 via the network.

[0158] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the claims that such modifications and improvements can also be included within the technical scope of the present invention.

[0159] It should be noted that the execution order of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before," "prior to," etc., and can be implemented in any order unless the output of a previous process is used in a subsequent process. Even if the operational flow in the claims, specifications, and drawings is described using "first," "next," etc. for convenience, this does not mean that the process must be implemented in this order. The expression "A and / or B" may mean "A, B, or A and C." The expression "A, B, and / or C" may mean "any one of A, B, and C, or any combination of two or more of these."

[0160] 10 Atomic structure estimation device 110 Generation unit 120 Acquisition unit 130 Global search unit 140 Local search unit 150 Output unit 160 Evaluation unit 170 Analysis unit 402 Virtual molecule 602 First candidate structure image 604 First candidate structure image 606 First candidate structure image 702 Basic candidate structure 802 Selected atomic structure 902 Three-dimensional structure model 904 Three-dimensional structure model 1102 Free state 1104 Open state 1106 Close state 1108 Condense state 1210 Trajectory 1222 Generated image 1224 Generated image 1226 Generated image

Claims

1. An apparatus for estimating the atomic structure of a biomolecule comprising: a generation unit that generates a generated image based on a target biomolecule; an acquisition unit that acquires a plurality of first candidate structure images based on two-dimensional projection images from a plurality of prepared atomic structures; a global search unit that selects, from the plurality of atomic structures, one whose structure and / or molecular orientation is highly similar to that of the target biomolecule as a basic candidate structure based on the similarity between the generated image and the first candidate structure image; and a local search unit that acquires a plurality of second candidate structure images based on two-dimensional projection images of a plurality of modified atomic structures obtained by modifying the basic candidate structure, and selects, from the modified atomic structures, one whose structure and / or molecular orientation is highly similar to that of the target biomolecule based on the similarity between the second candidate structure image and the generated image.

2. The atomic structure estimation device according to claim 1, wherein the acquisition unit acquires a plurality of first candidate structure images based on two-dimensional projection images projected from a plurality of orientations surrounding the atomic structure.

3. The atomic structure estimation device according to claim 1, wherein the acquisition unit acquires a first candidate structure image based on a two-dimensional projection image of the atomic structure reproduced in a molecular dynamics calculation.

4. The atomic structure estimation device according to claim 1, wherein the local search unit increases the degree of similarity to the target biomolecule by repeatedly modifying the basic candidate structure multiple times while maintaining the biomolecule structure.

5. The atomic structure estimation device according to claim 1, wherein the local search unit simulates the structural transition of the modified atomic structure by a cascade selection molecular dynamics calculation (PaCS-MD).

6. The atomic structure estimation device according to claim 2, wherein the global search unit matches a plurality of first candidate structure images with each of a plurality of generated images, and selects the basic candidate structure based on a similarity that is a matching result.

7. The atomic structure estimation device according to claim 6, wherein the global search unit selects the basic candidate structure based on the maximum, minimum, average, and / or median similarity of the multiple first candidate structure images matched to each of the multiple generated images.

8. The atomic structure estimation device according to claim 1, wherein the global search section calculates the similarity between the generated image and the first candidate structure image by parallel calculation.

9. The atomic structure estimation device according to claim 1, wherein the generation unit generates a generated image based on a particle image of the target biomolecule, the generation unit reduces noise in the particle image, and the generated image is a particle image with reduced noise.

10. The atomic structure estimation device according to claim 9, wherein the generation unit reduces noise in the particle image by generating an image using machine learning.

11. The atomic structure estimation device according to claim 10, wherein the generation unit applies a Fourier transform to the particle image with reduced noise to generate a target wavenumber space image which is the image after the Fourier transform as the generated image, and the acquisition unit applies a Fourier transform to a two-dimensional projection image of the atomic structure to acquire the wavenumber space image which is the image after the Fourier transform as the first candidate structure image.

12. The atom estimation device according to claim 11, wherein the generation unit removes a group of dust images in the latent space.

13. The atomic structure estimation device according to claim 1, further comprising an output unit that outputs a three-dimensional structure model of the modified atomic structure selected by the local search unit.

14. The atomic structure estimation device according to claim 1, further comprising an evaluation unit that evaluates and outputs interactions between fragments (residues) and / or between domains of the modified atomic structure selected by the local search unit.

15. The atomic structure estimation device according to claim 13, further comprising an analysis unit that analyzes structural features of the basic candidate structure and / or the modified atomic structure selected by the local search unit using principal component analysis, and the output unit outputs the results of the principal component analysis by plotting them on a plane or in space.

16. The atomic structure estimation device according to claim 15, wherein the global search unit assigns the generated image to one of the basic candidate structures based on the similarity, and the output unit further displays the number of attributes of the generated image to each of the basic candidate structures on the plane or the space.

17. The atomic structure estimation device according to claim 16, wherein the output unit displays the attribution numbers as a heat map on the plane or in the space.

18. The atomic structure estimation device according to claim 16, wherein the output unit displays the generated images belonging to the adjacent basic candidate structures on the plane or in the space in succession, thereby displaying the structural transition of the target biomolecule as a video.

19. A method for estimating the atomic structure of a biomolecule, comprising: a generation step of generating a generated image based on a target biomolecule; an acquisition step of acquiring a plurality of first candidate structure images based on two-dimensional projection images of a plurality of prepared atomic structures; a global search step of selecting, from the plurality of atomic structures, one having a high degree of similarity in structure and / or molecular orientation to the target biomolecule as a basic candidate structure based on the similarity between the generated image and the candidate structure image; and a local search step of acquiring a plurality of second candidate structure images based on two-dimensional projection images of a plurality of modified atomic structures obtained by modifying the basic candidate structure, and selecting, from the modified atomic structures, one having a high degree of similarity in structure and / or molecular orientation to the target biomolecule based on the similarity between the second candidate structure image and the generated image.

20. A program that is executed by a computer and causes the computer to function as: a generation unit that generates a generated image based on a target biomolecule; an acquisition unit that acquires a plurality of first candidate structure images based on two-dimensional projection images for a plurality of prepared atomic structures; a global search unit that selects, from the plurality of atomic structures, one whose structure and / or molecular orientation is highly similar to that of the target biomolecule as a basic candidate structure based on the similarity between the generated image and the candidate structure image; and a local search unit that acquires a plurality of second candidate structure images based on two-dimensional projection images of a plurality of modified atomic structures obtained by modifying the basic candidate structure, and selects, from the modified atomic structures, one whose structure and / or molecular orientation is highly similar to that of the target biomolecule based on the similarity between the second candidate structure image and the generated image.

Images