A method and system for cryo-em density map data compression and decompression

Abstract

The application discloses a cryo-EM density map data compression and decompression method and system, and belongs to the technical field of computer images, which comprises the following steps: obtaining cryo-EM density map data to be compressed; randomly sampling in the data coordinate range, and carrying out hash coding on the coordinates of the sampling points to obtain a feature vector of the coordinates; taking the feature vector with added uniform noise as the input of a neural network, training the neural network by minimizing the error between the predicted value of the network output and the original data value at the sampling point, and obtaining optimized neural network parameters; quantizing the feature vector, and compressing the quantized feature vector and the neural network parameters by using a lossless coding method to obtain compressed data. The decompression method is the inverse process of the compression method. The application can realize ultra-high compression ratio while reducing compression loss as much as possible, and supports on-demand decompression and access of data.

Description

Technical Field

[0001] This invention belongs to the field of computer image technology, and in particular relates to a method and system for compressing and decompressing cryo-electron microscopy density map data. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Cryo-electron microscopy density maps are three-dimensional electron density distribution data of biological macromolecules or cell structures obtained through cryo-electron microscopy. They are reconstructed from two-dimensional projection images of rapidly frozen samples and can reflect the shape, composition, and interactions of biomolecules. Typically, density maps have resolutions ranging from a few nanometers to near-atomic levels. They can be used to resolve the structures of proteins, nucleic acids, and other biological components, aiding in drug design and cell research. Their generation relies on high-quality samples and advanced image processing techniques, making them an important tool in modern structural biology.

[0004] With the widespread application of cryo-electron microscopy (cryo-EM) technology, the amount of cryo-EM density map data in publicly available databases is growing rapidly. Simultaneously, as cryo-EM technology continues to advance and imaging resolution increases, the size of individual cryo-EM density map data points is also constantly increasing. Therefore, the large volume of cryo-EM density map data poses significant challenges to both storage and transmission.

[0005] Currently, compression technology offers a potential solution to the problem of excessively large data volumes. While existing lossless compression techniques can effectively preserve original information, their compression effects are quite limited, typically achieving compression ratios of only 1.5-3 times. Furthermore, their algorithmic performance is nearing its theoretical limit, making further improvement difficult and failing to fundamentally address the issue of excessively large cryo-electron microscopy (cryo-EM) density map data volumes. Traditional lossy compression techniques, while achieving higher compression ratios, often incur compression losses. Because cryo-EM density map data is filled with significant noise, these compression losses are more pronounced. Moreover, data compressed and then decompressed using traditional lossy compression methods exhibits noticeable visual artifacts during visualization. Additionally, traditional compression methods often require complete decompression of all data when accessing it, failing to support on-demand decompression and access, resulting in substantial resource waste when only a portion of the data is needed. Summary of the Invention

[0006] To address the shortcomings of the existing technologies, this invention provides a method and system for compressing and decompressing cryo-electron microscopy density map data. By employing a deep neural network based on multi-resolution hash coding to represent the cryo-electron microscopy density map data, and introducing quantization and entropy coding techniques to efficiently compress the data, the invention achieves an ultra-high compression ratio while minimizing compression loss, and supports on-demand decompression and access of the data.

[0007] In a first aspect, the present invention provides a method for compressing cryo-electron microscopy density map data.

[0008] A method for compressing cryo-electron microscopy density map data includes:

[0009] Obtain cryo-electron microscopy density map data to be compressed;

[0010] Randomly sample within the data coordinate range, hash the coordinates of the sampled points to obtain the feature vector of the coordinates;

[0011] Using feature vectors with added uniform noise as input to the neural network, the neural network is trained by minimizing the error between the network output prediction value and the original data value at the sampling point, thus obtaining the optimized neural network parameters;

[0012] The feature vectors are quantized, and lossless encoding methods are used to compress the quantized feature vectors and neural network parameters to obtain compressed data.

[0013] A further technical solution is that the hash encoding adopts multi-resolution hash encoding, which uses multi-resolution hash encoding to encode the coordinates of the sampling points, including:

[0014] Create a set of multi-resolution grids. For each resolution grid, map and calculate the vertex information of the grid where the sampling point coordinates are located, and then store it in a hash table.

[0015] Based on the hash table, generate feature vectors of the sampling point coordinates;

[0016] The mapping calculation is as follows: when the number of vertices in the grid is less than the maximum length of the hash table, a one-to-one mapping relationship is constructed between the coordinates of each grid vertex and the hash table entry, as follows:

[0017] ;

[0018] In the above formula, This represents the coordinates of a grid vertex in the grid coordinate system. For the first i Level resolution grid resolution, Indicates the first in the hash table i One hash table entry;

[0019] When the number of vertices in the mesh is not less than the maximum length of the hash table, a mapping from mesh vertex coordinates to hash table indices is constructed using a hash function, which is:

[0020] ;

[0021] In the above formula, , and Represent three distinct arbitrary prime numbers. This represents the XOR operation. This represents the modulo operation. This indicates the maximum length of the hash table.

[0022] A further technical solution involves, for each sampling point, identifying the nearest multiple grid vertices around that sampling point in each grid level and retrieving the corresponding feature vectors from a hash table. Then, using trilinear interpolation, the feature vector corresponding to that sampling point in that grid level is calculated. Uniform noise is then added to all the obtained feature vectors. To simulate quantization error; among which, Indicates the length of the quantization interval.

[0023] A further technical solution involves quantizing the feature vectors stored in the hash table, as follows:

[0024] ;

[0025] In the above formula, This is the value before quantization. It is the quantized value. Indicates normalization. This indicates rounding.

[0026] Secondly, the present invention provides a method for decompressing cryo-electron microscopy density map data.

[0027] A method for decompressing cryo-electron microscopy density map data includes:

[0028] Obtain compressed cryo-electron microscopy density map data using the cryo-electron microscopy density map data compression method proposed in the first aspect;

[0029] The compressed data is decompressed using a lossless encoding method to obtain quantized feature vectors and neural network parameters. The quantized feature vectors are then dequantized and stored in a hash table, and the neural network is restored.

[0030] Based on the coordinates corresponding to the data to be decompressed, the data is hashed and then input into the restored neural network to obtain the density value at each coordinate point, thus completing the data decompression.

[0031] A further technical solution involves inverse quantization of the quantized feature vectors, as follows:

[0032] ;

[0033] In the above formula, This is the value before quantization. It is the quantized value. Indicates the length of the quantization interval. and These are the maximum and minimum values ​​of the eigenvectors before quantization, respectively.

[0034] Thirdly, the present invention provides a cryo-electron microscopy density map data compression system.

[0035] A cryo-electron microscopy density map data compression system, comprising:

[0036] The data acquisition module is used to acquire the cryo-electron microscopy density map data to be compressed;

[0037] The data encoding module is used to randomly sample within the data coordinate range, hash-encode the coordinates of the sampled points, and obtain the feature vector of the coordinates.

[0038] The neural network training module is used to train the neural network by taking the feature vector with uniform noise added as the input of the neural network, and by minimizing the error between the network output prediction value and the original data value at the sampling point, so as to obtain the optimized neural network parameters.

[0039] The compression module is used to quantize the feature vectors and compress the quantized feature vectors and neural network parameters using a lossless encoding method to obtain compressed data.

[0040] Fourthly, the present invention provides a cryo-electron microscopy density map data decompression system.

[0041] A cryo-electron microscopy density map data decompression system, comprising:

[0042] The compressed data acquisition module is used to acquire the compressed cryo-electron microscopy density map data obtained by compressing it using the cryo-electron microscopy density map data compression method proposed in the first aspect.

[0043] The neural network reconstruction module is used to decompress compressed data using a lossless encoding method, obtain quantized feature vectors and neural network parameters, dequantize the quantized feature vectors and store them in a hash table, and reconstruct the neural network.

[0044] The decompression module is used to input the coordinates corresponding to the data to be decompressed into the restored neural network after hash encoding, so as to obtain the density value at each coordinate point and complete the data decompression.

[0045] Fifthly, the present invention also provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-described cryo-electron microscopy density map data compression method or cryo-electron microscopy density map data decompression method when executing the executable instructions stored in the memory.

[0046] In a sixth aspect, the present invention also provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-described cryo-electron microscopy density map data compression method or cryo-electron microscopy density map data decompression method.

[0047] The above one or more technical solutions have the following beneficial effects:

[0048] 1. This invention proposes a method and system for compressing and decompressing cryo-electron microscopy density map data. A deep neural network based on multi-resolution hash coding is designed. By training this deep neural network, the network fits the cryo-electron microscopy density map data, transforming the data into a form represented by neural network parameters. Quantization and entropy coding techniques are then introduced to efficiently compress the trained neural network parameter data, achieving an ultra-high compression ratio while minimizing compression loss. Furthermore, during decompression, the quantized and entropy-coded network parameters are restored, and the corresponding cryo-point density data can be decompressed according to specified coordinates or coordinate ranges, enabling on-demand decompression and access of the data.

[0049] 2. In the cryo-electron microscopy density map data compression method based on implicit neural representation and quantization proposed in this invention, in order to reduce the subsequent data quantization loss, a simulated quantization method is also introduced in the neural network training process. That is, by adding uniform noise to the feature vector used for storage hash encoding, the quantization loss is simulated, thereby reducing the quantization error in the quantization stage during neural network optimization, thereby improving the accuracy and reliability of the density data output by the neural network.

[0050] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0051] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0052] Figure 1 This is an overall flowchart of the cryo-electron microscopy density map data compression method described in this embodiment of the invention;

[0053] Figure 2This is a visualization result of cryo-electron microscopy density map data before and after compression and decompression in an embodiment of the present invention; wherein, ( a ) is the visualization result of the original data, b This is a magnified view of a local region A in the original data. c This is a visualization of the decompressed data. d () is a magnified view of a local area A in the decompressed data. Detailed Implementation

[0054] It should be noted that the following detailed descriptions are exemplary and are intended only to describe specific embodiments and to provide further explanation of the invention, and are not intended to limit the scope of exemplary embodiments of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0055] Example 1

[0056] This embodiment provides a method for compressing cryo-electron microscopy density map data, such as... Figure 1 As shown, it mainly includes the following steps:

[0057] Step S1: Obtain the cryo-electron microscopy density map data to be compressed. The obtained density map data is three-dimensional electron density distribution data.

[0058] Step S2: Randomly sample within the data coordinate range, hash the coordinates of the sampled points, and obtain the feature vector of the coordinates.

[0059] Step S3: Using the feature vector with added uniform noise as the input to the neural network, train the neural network by minimizing the error between the network output prediction value and the original data value at the sampling point to obtain the optimized neural network parameters.

[0060] In this embodiment, a neural network is used to fit the cryo-electron microscopy density map data. The fitting process is the neural network training process described in steps S2 and S3 above. Through training, the density map data is fitted into neural network parameter data. The process is as follows:

[0061] First, within the coordinate range of the density map data, a set of coordinate points are randomly generated as sampling points, and the coordinates of the sampling points are multi-resolution hash encoded. The encoded coordinates are represented by a set of feature vectors.

[0062] Specifically, multi-resolution hashing is a location encoding method. It uses multi-resolution hashing to encode the location of sampling point coordinates. For a set of data to be compressed, a multi-resolution grid is created. ,in and Let represent the sparsest and densest grids, respectively, then the grid resolution is... satisfy:

[0063] ;

[0064] in, , N Indicates the number of multi-resolution grids.

[0065] For multi-resolution meshes, a hash table is used to store vertex coordinate information. This involves mapping and calculating the vertex information of the mesh containing the sampled point's coordinates, then storing this information in the hash table. Only the mesh vertex information is mapped and stored during this process; the sampled point information can be obtained through trilinear interpolation based on its surrounding vertices. The maximum length of the hash table is [value missing]. When the number of vertices in the mesh is less than At that time, a one-to-one mapping relationship is constructed between the coordinates of each grid vertex and the hash table entry, as follows:

[0066] ;

[0067] In the above formula, This represents the coordinates of a grid vertex in the grid coordinate system. For the first i Level resolution grid resolution, Indicates the first in the hash table i Each hash table entry.

[0068] When the number of grid vertices is greater than At that time, the mapping from vertex coordinates to hash table indices is established using the following hash function:

[0069] ;

[0070] In the above formula, , and Represent three distinct arbitrary prime numbers. This represents the XOR operation. This indicates the modulo operation.

[0071] Furthermore, for each hash table entry, a value of length is stored. Based on the hash table obtained from the above mapping, feature vectors of sampling point coordinates are generated. In the latter case, multiple grid vertices may map to the same hash table entry, resulting in a hash collision. However, since multi-resolution hash encoding in this embodiment serves the subsequent neural network, the situation where multiple vertices map to the same hash table entry can be regarded as a weighted average. The neural network can automatically optimize the weights through gradient descent. Therefore, it is not necessary to explicitly handle the hash collision; instead, it can be left to the subsequent neural network for automatic optimization.

[0072] For each sampling point, find the eight nearest grid vertices around that sampling point in each grid level and obtain the corresponding feature vectors from the hash table. Then, use trilinear interpolation to calculate the feature vector corresponding to that sampling point in that grid level. Add uniform noise to all the feature vectors obtained above. To simulate quantization error, where This represents the length of the quantization interval, typically taken as... By adding uniform noise to the feature vector used for storage hash encoding, the error caused in the actual quantization process can be simulated better. Furthermore, the process of adding uniform noise is differentiable, which allows the neural network to consider quantization loss during training, thereby reducing the error caused in the actual quantization process.

[0073] Next, the noisy feature vector is input into a neural network consisting of two linear layers, such as an MLP (Multilayer Perceptron) neural network, to obtain the predicted value output by the neural network. This predicted value is the density value of the predicted sampling point location. In summary, this embodiment uses the MSE loss function to compare the predicted value with the density value of the original data at the sampling point to obtain the loss, which can be expressed as:

[0074] ;

[0075] in, This represents neural network prediction data. This represents the raw data, all of which can be used as coordinates. The function representation, and then Coordinates representing the output of the neural network Predicted data at the location, Representing coordinates The original data at that location.

[0076] Based on the above loss function, the network parameters are optimized by backpropagation gradient, and after several iterations, the optimized neural network and its parameters are finally obtained.

[0077] Step S4: Quantize the feature vector and compress the quantized feature vector and neural network parameters using a lossless encoding method to obtain compressed data.

[0078] After network training is complete, the feature vectors stored in the hash table of hash encoding are first quantized, which can be represented as:

[0079] ;

[0080] In the above formula, This is the value before quantization. It is the quantized value. Indicates normalization. This indicates rounding.

[0081] Secondly, the quantized feature vectors and the remaining neural network parameters are further compressed using the LZMA2 lossless coding method (Lempel-Ziv-Markov chain Algorithm, an algorithm for lossless data compression) or other commonly used entropy coding and other lossless coding methods to obtain the final compressed data, thus completing the data compression.

[0082] Since the trained neural network can approximate the original data and the number of parameters of the neural network is much smaller than the size of the original data, the above process can be regarded as a lossy compression. Moreover, this method uses quantization and lossless coding to further compress the parameters of the neural network, ultimately achieving an ultra-high compression ratio.

[0083] Example 2

[0084] This embodiment proposes a method for decompressing cryo-electron microscopy density map data. This decompression method is the inverse process of the compression method, and specifically includes the following steps:

[0085] Step S1: Obtain compressed data of cryo-electron microscopy density map, which is the compressed data obtained in Example 1 above.

[0086] Step S2: Decompress the compressed data using a lossless encoding method to obtain the quantized feature vector and neural network parameters. Dequantize the quantized feature vector and store it in a hash table. Then, reconstruct the neural network based on the previously saved neural network parameters.

[0087] The inverse quantization of the quantized feature vector is as follows:

[0088] ;

[0089] In the above formula, This is the value before quantization. It is the quantized value. Indicates the length of the quantization interval. and These are the maximum and minimum values ​​of the eigenvectors before quantization, respectively.

[0090] Step S3: Based on the coordinates corresponding to the data to be decompressed, the data is hashed and then input into the restored neural network to obtain the density value at each coordinate point, thus completing the data decompression.

[0091] Specifically, to decompress all data, coordinates need to be generated for each point on the original cryo-electron microscopy density map data. Based on the decompressed hash table, these coordinates are hash-encoded and input into the reconstructed neural network to obtain the density value corresponding to each point output by the neural network, thus completing the data decompression. To decompress only a portion of the data, coordinates corresponding to the required portion (i.e., the data to be decompressed) need to be generated. Based on the decompressed hash table, these coordinates are hash-encoded and input into the neural network to obtain the density value at each coordinate point, thus completing the decompression of that portion of the data. Through this method, on-demand access and decompression of compressed data can be achieved.

[0092] Furthermore, by using the compression and decompression method proposed in this embodiment to experiment with cryo-electron microscopy density maps, the following results can be obtained: Figure 2 The visualization results of the data before and after compression and decompression are shown.

[0093] Example 3

[0094] This embodiment provides a cryo-electron microscopy density map data compression system, including:

[0095] The data acquisition module is used to acquire the cryo-electron microscopy density map data to be compressed;

[0096] The data encoding module is used to randomly sample within the data coordinate range, hash-encode the coordinates of the sampled points, and obtain the feature vector of the coordinates.

[0097] The neural network training module is used to train the neural network by taking the feature vector with uniform noise added as the input of the neural network, and by minimizing the error between the network output prediction value and the original data value at the sampling point, so as to obtain the optimized neural network parameters.

[0098] The compression module is used to quantize the feature vectors and compress the quantized feature vectors and neural network parameters using a lossless encoding method to obtain compressed data.

[0099] Example 4

[0100] This embodiment provides a cryo-electron microscopy density map data decompression system, including:

[0101] The compressed data acquisition module is used to acquire compressed data of cryo-electron microscopy density maps, which is the compressed data obtained in Embodiment 1 above.

[0102] The neural network reconstruction module is used to decompress compressed data using a lossless encoding method, obtain quantized feature vectors and neural network parameters, dequantize the quantized feature vectors and store them in a hash table, and reconstruct the neural network.

[0103] The decompression module is used to input the coordinates corresponding to the data to be decompressed into the restored neural network after hash encoding, so as to obtain the density value at each coordinate point and complete the data decompression.

[0104] Example 5

[0105] This embodiment provides an electronic device, including: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the method provided in this embodiment.

[0106] Example 6

[0107] This embodiment also provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, will cause the processor to execute the method described above in this embodiment.

[0108] The steps involved in Embodiments 3 to 6 above correspond to Method Embodiments 1 and 2, respectively. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0109] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0110] The above description is only a preferred embodiment of the present invention. Although the specific implementation of the present invention has been described in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the present invention.

Claims

1. A method for compressing cryo-electron microscopy density map data, characterized in that, include: Obtain cryo-electron microscopy density map data to be compressed; The acquired density map data is three-dimensional electron density distribution data; Randomly sample within the data coordinate range, hash the coordinates of the sampled points to obtain the feature vector of the coordinates; The hash encoding employs multi-resolution hash encoding, which encodes the coordinates of the sampling points using multi-resolution hash encoding, including: Create a set of multi-resolution grids. For each resolution grid, map and calculate the vertex information of the grid where the sampling point coordinates are located, and then store it in a hash table. Based on the hash table, generate feature vectors of the sampling point coordinates; The mapping calculation is as follows: when the number of vertices in the grid is less than the maximum length of the hash table, a one-to-one mapping relationship is constructed between the coordinates of each grid vertex and the hash table entry, as follows: ； In the above formula, This represents the coordinates of a grid vertex in the grid coordinate system. For the first i Level resolution grid resolution, Indicates the first in the hash table i One hash table entry; When the number of vertices in the mesh is not less than the maximum length of the hash table, a mapping from mesh vertex coordinates to hash table indices is constructed using a hash function, which is: ； In the above formula, , and Represent three distinct arbitrary prime numbers. This represents the XOR operation. This represents the modulo operation. Indicates the maximum length of the hash table; Using feature vectors with added uniform noise as input to the neural network, the neural network is trained by minimizing the error between the network output prediction value and the original data value at the sampling point, thus obtaining the optimized neural network parameters; For each sampling point, the nearest multiple grid vertices around the sampling point in each grid level are determined, and the corresponding feature vectors are obtained from a hash table. Then, the trilinear interpolation method is used to calculate the feature vector corresponding to the sampling point in that grid level. Uniform noise is added to all the obtained feature vectors. To simulate quantization error, Indicates the length of the quantization interval; The loss is calculated by comparing the predicted value with the density value of the original data at the sampling point using the MSE loss function, and is expressed as: ; in, This represents neural network prediction data. This represents the raw data, all of which can be used as coordinates. The function representation, Coordinates representing the output of the neural network The predicted data at that location, i.e., the predicted density values ​​at the sampling point locations. Representing coordinates The original data at the location; The feature vectors are quantized, and the quantized feature vectors and neural network parameters are compressed using a lossless encoding method to obtain compressed data. The quantization of the feature vectors stored in the hash table is as follows: ； In the above formula, This is the value before quantization. It is the quantized value. Indicates normalization. This indicates rounding. Indicates the length of the quantization interval.

2. A method for decompressing cryo-electron microscopy density map data, characterized in that, include: Obtain compressed cryo-electron microscopy density map data using the cryo-electron microscopy density map data compression method described in claim 1; The compressed data is decompressed using a lossless encoding method to obtain quantized feature vectors and neural network parameters. The quantized feature vectors are then dequantized and stored in a hash table, and the neural network is restored. Based on the coordinates corresponding to the data to be decompressed, the data is hashed and then input into the restored neural network to obtain the density value at each coordinate point, thus completing the data decompression.

3. The method for decompressing cryo-electron microscopy density map data as described in claim 2, characterized in that, Dequantizing the quantized feature vector yields: ； In the above formula, This is the value before quantization. It is the quantized value. Indicates the length of the quantization interval. and These are the maximum and minimum values ​​of the eigenvectors before quantization, respectively.

4. A cryo-electron microscopy density map data compression system, characterized in that, include: The data acquisition module is used to acquire the cryo-electron microscopy density map data to be compressed. The acquired density map data is three-dimensional electron density distribution data. The data encoding module is used to randomly sample within the data coordinate range, hash-encode the coordinates of the sampled points, and obtain the feature vector of the coordinates. The hash encoding employs multi-resolution hash encoding, which encodes the coordinates of the sampling points using multi-resolution hash encoding, including: Create a set of multi-resolution grids. For each resolution grid, map and calculate the vertex information of the grid where the sampling point coordinates are located, and then store it in a hash table. Based on the hash table, generate feature vectors of the sampling point coordinates; The mapping calculation is as follows: when the number of vertices in the grid is less than the maximum length of the hash table, a one-to-one mapping relationship is constructed between the coordinates of each grid vertex and the hash table entry, as follows: ； In the above formula, This represents the coordinates of a grid vertex in the grid coordinate system. For the first i Level resolution grid resolution, Indicates the first in the hash table i One hash table entry; When the number of vertices in the mesh is not less than the maximum length of the hash table, a mapping from mesh vertex coordinates to hash table indices is constructed using a hash function, which is: ； In the above formula, , and Represent three distinct arbitrary prime numbers. This represents the XOR operation. This represents the modulo operation. Indicates the maximum length of the hash table; The neural network training module uses feature vectors with added uniform noise as input to the neural network. It trains the network by minimizing the error between the predicted output value and the original data value at the sampling point, thus obtaining optimized neural network parameters. Specifically, for each sampling point, it determines the nearest multiple grid vertices in each grid level and retrieves the corresponding feature vectors from a hash table. Then, it uses trilinear interpolation to calculate the feature vector corresponding to that sampling point in that grid level. Uniform noise is added to all the obtained feature vectors. To simulate quantization error; among which, Indicates the length of the quantization interval; The loss is calculated by comparing the predicted value with the density value of the original data at the sampling point using the MSE loss function, and is expressed as: ; in, This represents neural network prediction data. This represents the raw data, all of which can be used as coordinates. The function representation, Coordinates representing the output of the neural network The predicted data at that location, i.e., the predicted density values ​​at the sampling point locations. Representing coordinates The original data at the location; The compression module is used to quantize the feature vectors and compress the quantized feature vectors and neural network parameters using a lossless encoding method to obtain compressed data. The quantization of the feature vectors stored in the hash table is as follows: ； In the above formula, This is the value before quantization. It is the quantized value. Indicates normalization. This indicates rounding. Indicates the length of the quantization interval.

5. A cryo-electron microscopy density map data decompression system, characterized in that, include: The compressed data acquisition module is used to acquire the compressed cryo-electron microscopy density map data obtained by compressing it using the cryo-electron microscopy density map data compression method described in claim 1. The neural network reconstruction module is used to decompress compressed data using a lossless encoding method, obtain quantized feature vectors and neural network parameters, dequantize the quantized feature vectors and store them in a hash table, and reconstruct the neural network. The decompression module is used to input the coordinates corresponding to the data to be decompressed into the restored neural network after hash encoding, so as to obtain the density value at each coordinate point and complete the data decompression.

6. A computer-readable storage medium, characterized in that, The device stores executable instructions that, when executed by a processor, implement the cryo-electron microscopy density map data compression method of claim 1 or the cryo-electron microscopy density map data decompression method of any one of claims 2-3.

7. A computer program product, characterized in that, The computer program product includes executable instructions stored in a computer-readable storage medium; When the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, it implements the cryo-electron microscopy density map data compression method of claim 1 or the cryo-electron microscopy density map data decompression method of any one of claims 2-3.

Images