Model training method, spatial metabolic data acquisition method and device, and storage medium

Abstract

This disclosure provides a model training method, a spatial metabolic data acquisition method and apparatus, and a storage medium. The model training method includes: registering mass spectrometry image samples and stained image samples of biological tissue slices to obtain target image samples; acquiring spatial metabolic data of each point in a specified point set in the target image samples; processing the spatial metabolic data of each point using the encoder of a generative adversarial network (GAN) model to obtain predicted coordinates of each point; processing the predicted coordinates of each point using the decoder of a GAN model to obtain spatial metabolic reconstruction data of each point; determining the confidence level of the coordinates and the confidence level of the predicted coordinates of each point using the discriminator of a GAN model; determining a loss value based on the spatial metabolic data, spatial metabolic reconstruction data, coordinates, predicted coordinates, confidence level of the coordinates, and confidence level of the predicted coordinates of each point; and training the encoder, decoder, and discriminator using the loss value.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a model training method, a method and apparatus for acquiring spatial metabolic data, and a storage medium. Background Technology

[0002] Traditional single-omics techniques (e.g., genomics, transcriptomics, metabolomics, or proteomics) can only capture information at the single molecular level, making it difficult to comprehensively reveal the complex interactions between gene regulation and metabolic networks in the tissue microenvironment. With the rise of spatial multi-omics technologies, researchers hope to simultaneously analyze gene expression and metabolic activities while preserving the spatial structure of tissues, thereby providing a multidimensional perspective for a deeper understanding of the heterogeneity of the tissue microenvironment and the mechanisms of disease development.

[0003] For example, spatial transcriptomics (ST) uses in-situ sequencing (ISS) or in-situ hybridization (ISH) techniques to precisely locate gene expression in tissue sections. Spatial metabolomics (SM) combines mass spectrometry imaging (MSI) to directly detect and locate metabolites in two-dimensional or three-dimensional space within cells or tissues. By integrating spatial transcriptomics and spatial metabolomics data, data continuity is enhanced, enabling multi-omics and multimodal joint analysis. Summary of the Invention

[0004] The inventors noted that, in related technologies, with the rapid development of spatial multimodal analysis technology, the SMA (Spatial Multimodal Analysis protocol) has achieved simultaneous and accurate measurement of spatial transcriptomics and MSI on a single tissue slice, preserving the specificity and sensitivity of both modalities. However, the spatial coordinate information of the spatial transcriptomics and spatial metabolomics data generated by this method is still not in the same reference coordinate system. Therefore, in the process of integrating spatial transcriptomics and spatial metabolomics data, precise point-to-point alignment of the spatial transcriptomics and spatial metabolomics data remains a challenge.

[0005] Accordingly, this disclosure provides a model training method that uses a generative adversarial network model to learn the correlation between the coordinates of each point and the spatial metabolic data of each point. This method can accurately predict the spatial metabolic data of the target point, so as to achieve precise point-to-point alignment of spatial transcriptome data and spatial metabolome data, and make the integration between spatial transcriptome and spatial metabolome not limited by spatial resolution differences.

[0006] In a first aspect of this disclosure, a model training method is provided, comprising: performing mass spectrometry imaging on a biological tissue slice sample to obtain a mass spectrometry image sample; performing HE staining on the biological tissue slice sample to obtain a stained image sample; registering the mass spectrometry image sample with the stained image sample to obtain a target image sample; acquiring spatial metabolic data of each point in a specified point set in the target image sample; processing the spatial metabolic data of each point using an encoder in a generative adversarial network model to obtain predicted coordinates of each point; and processing the predicted coordinates of each point using a decoder in the generative adversarial network model to obtain the predicted coordinates of each point. Spatial metabolic reconstruction data for each point; the probability that the coordinates of each point are true coordinates is identified by the discriminator in the generative adversarial network model, which is used as the confidence level of the coordinates of each point; the probability that the predicted coordinates of each point are true coordinates is identified by the discriminator, which is used as the confidence level of the predicted coordinates of each point; a loss value is determined based on the spatial metabolic data of each point, the spatial metabolic reconstruction data of each point, the coordinates of each point, the predicted coordinates of each point, the confidence level of the coordinates of each point, and the confidence level of the predicted coordinates of each point; the encoder, the decoder, and the discriminator are trained using the loss value.

[0007] In some embodiments, determining the loss value includes: determining a first sub-loss value based on the spatial metabolic data and the spatial metabolic reconstruction data of each point; determining a second sub-loss value based on the coordinates of each point, the predicted coordinates of each point, the confidence level of the coordinates of each point, and the confidence level of the predicted coordinates of each point; and obtaining the loss value based on the first sub-loss value and the second sub-loss value.

[0008] In some embodiments, determining the first sub-loss value includes: calculating the weighted mean square error of the spatial metabolic data and the spatial metabolic reconstruction data at each point to obtain a first deviation value; calculating the mean absolute error of the spatial metabolic data and the spatial metabolic reconstruction data at each point to obtain a second deviation value; and calculating the weighted sum of the first deviation value and the second deviation value to obtain the first sub-loss value.

[0009] In some embodiments, the spatial metabolic data of the i-th point is a data vector comprising M elements, wherein the j-th element of the data vector represents the metabolic intensity of the i-th point at the j-th mass-to-charge ratio. N is the total number of points in the specified point set. M represents the total number of mass-to-charge ratios.

[0010] In some embodiments, determining the weight of the first deviation value includes: constructing a metabolic intensity matrix, wherein the metabolic intensity matrix includes spatial metabolic data of N points in the specified point set; calculating the sum of all metabolic intensities of M mass-to-charge ratio values ​​in the metabolic intensity matrix to obtain a matrix sum; calculating the sum of all metabolic intensities of the j-th mass-to-charge ratio value in the metabolic intensity matrix to obtain a j-th sum; calculating the ratio of the j-th sum to the matrix sum to obtain a j-th ratio; and calculating the difference between a predetermined value and the j-th ratio to obtain the j-th weight value in the weight of the first deviation value.

[0011] In some embodiments, determining the second sub-loss value includes: calculating the mean absolute error between the coordinates of each point and the predicted coordinates of each point to obtain a coordinate error value; calculating the average confidence level of the coordinates of each point to obtain a first confidence level mean; calculating the average confidence level of the predicted coordinates of each point to obtain a second confidence level mean; calculating the weighted sum of the coordinate error value and the first confidence level mean to obtain an intermediate loss value; and calculating the difference between the intermediate loss value and the second confidence level mean to obtain the second sub-loss value.

[0012] In some embodiments, obtaining the loss value based on the first sub-loss value and the second sub-loss value includes: calculating a weighted sum of the first sub-loss value and the second sub-loss value to obtain the loss value.

[0013] In some embodiments, multiple addition points are selected within the neighborhood of each point in the specified point set; the spatial metabolic data of the point in the specified point set closest to the addition point is used as the spatial metabolic data of the addition point; the spatial metabolic data of the kth addition point is updated using the spatial metabolic data of the 8 neighboring points of the kth addition point. K is the total number of points to be added to the specified point set.

[0014] In some embodiments, selecting multiple add points in the neighborhood of each point in the specified point set includes: establishing a polar coordinate system for each point, wherein each point is the pole of the polar coordinate system, and a ray extending from the pole along a predetermined direction is the polar axis of the polar coordinate system; and taking the corresponding points of multiple candidate coordinates in the polar coordinate system as multiple add points.

[0015] In some embodiments, the plurality of candidate coordinates includes a first candidate coordinate. Second candidate coordinates and the third candidate coordinates , where R is the predetermined distance.

[0016] In some embodiments, registering the mass spectrometry image sample with the staining image sample to obtain a target image sample includes: extracting spatial metabolic data with a highly spatially variable mass-to-charge ratio from the mass spectrometry image sample to generate an image sample to be processed; and using an affine transformation matrix to map each pixel in the image sample to be processed to the staining image coordinate system to obtain the target image sample.

[0017] In some embodiments, a primitive transformation matrix is ​​trained to obtain the affine transformation matrix, wherein training the primitive transformation matrix includes: generating the primitive transformation matrix based on a horizontal scaling factor, a vertical scaling factor, a horizontal shearing factor, a vertical shearing factor, a horizontal translation, and a vertical translation; using the primitive transformation matrix, transforming the first coordinates of each pixel in the image sample to be processed into the stained image sample to obtain the second coordinates of each pixel; extracting the coordinates of feature points in the stained image sample that have a one-to-one correspondence with each pixel; calculating the mean square error of the coordinates of the second coordinates of each pixel and the coordinates of the feature points that correspond one-to-one with each pixel; and training the horizontal scaling factor, the vertical scaling factor, the horizontal shearing factor, the vertical shearing factor, the horizontal translation, and the vertical translation based on the mean square error of the coordinates to obtain the affine transformation matrix.

[0018] In a second aspect of this disclosure, a model training apparatus is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute instructions stored in the memory to implement the model training method as described in any of the above embodiments.

[0019] In a third aspect of this disclosure, a method for acquiring spatial metabolic data is provided, comprising: performing mass spectrometry imaging on a biological tissue section to be processed to obtain a mass spectrometry image; performing HE staining on the biological tissue section to be processed to obtain a staining image; registering the mass spectrometry image and the staining image to obtain a target image; acquiring the coordinates of each of a plurality of measurement points in the target image; and processing the coordinates of each measurement point using a decoder to obtain spatial metabolic data of each measurement point, wherein the decoder is trained using the model training method involved in any of the above embodiments.

[0020] In some embodiments, registering the mass spectrometry image with the staining image to obtain a target image includes: extracting spatial metabolic data with spatially highly variable mass-to-charge ratio values ​​from the mass spectrometry image to generate an image to be processed; and using an affine transformation matrix to map each pixel in the image to be processed to the coordinate system of the staining image to obtain the target image.

[0021] In a fourth aspect of this disclosure, a spatial metabolic data acquisition apparatus is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute instructions stored in the memory to implement the spatial metabolic data acquisition method as described in any of the above embodiments.

[0022] In a fifth aspect of this disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method as described in any of the above embodiments.

[0023] In a sixth aspect of this disclosure, a computer program product is provided, including computer instructions, wherein the computer instructions, when executed by a processor, implement the method as described in any of the above embodiments.

[0024] Other features and advantages of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic flowchart of a model training method according to an embodiment of the present disclosure;

[0027] Figure 2 This is a schematic diagram of the image registration process according to an embodiment of the present disclosure;

[0028] Figure 3 This is a schematic diagram of a high-density spatial metabolic data generation method according to an embodiment of the present disclosure;

[0029] Figure 4 This is a schematic diagram of a high-density spatial metabolic data generation method according to another embodiment of this disclosure;

[0030] Figure 5 This is a schematic diagram of a high-density spatial metabolic data generation method according to yet another embodiment of this disclosure;

[0031] Figure 6 This is a schematic diagram of the structure of a generative adversarial network model according to an embodiment of the present disclosure;

[0032] Figure 7 This is a schematic diagram of the structure of a model training apparatus according to an embodiment of the present disclosure;

[0033] Figure 8 This is a flowchart illustrating a spatial metabolic data acquisition method according to an embodiment of the present disclosure;

[0034] Figure 9 This is a flowchart illustrating a spatial metabolic data acquisition method according to another embodiment of the present disclosure;

[0035] Figure 10 This is a schematic diagram of a high-density spatial metabolic data generation method according to yet another embodiment of this disclosure;

[0036] Figure 11 This is a schematic diagram of a high-density spatial metabolic data generation method according to yet another embodiment of this disclosure;

[0037] Figure 12 This is a flowchart illustrating a spatial metabolic data acquisition method according to yet another embodiment of this disclosure;

[0038] Figure 13 This is a schematic diagram of benchmark test results according to an embodiment of the present disclosure;

[0039] Figure 14 This is a schematic diagram of benchmark test results for another embodiment of this disclosure;

[0040] Figure 15 This is a schematic diagram of benchmark test results for yet another embodiment of this disclosure;

[0041] Figure 16 This is a schematic diagram of benchmark test results for yet another embodiment of this disclosure;

[0042] Figure 17 This is a schematic diagram of benchmark test results for yet another embodiment of this disclosure;

[0043] Figure 18 This is a schematic diagram of benchmark test results for yet another embodiment of this disclosure;

[0044] Figure 19 This is a schematic diagram of the experimental results of multi-omics data integration according to an embodiment of this disclosure;

[0045] Figure 20 This is a schematic diagram of the experimental results of multi-omics data integration according to another embodiment of this disclosure;

[0046] Figure 21This is a schematic diagram of the experimental results of multi-omics data integration according to yet another embodiment of this disclosure;

[0047] Figure 22 This is a schematic diagram of the experimental results of multi-omics data integration according to yet another embodiment of this disclosure;

[0048] Figure 23 This is a schematic diagram of the experimental results of multi-omics data integration according to yet another embodiment of this disclosure;

[0049] Figure 24 This is a schematic diagram of the experimental results of multi-omics data integration according to yet another embodiment of this disclosure;

[0050] Figure 25 This is a schematic diagram of clustering experiment results according to an embodiment of this disclosure;

[0051] Figure 26 This is a schematic diagram of clustering experiment results according to an embodiment of this disclosure;

[0052] Figure 27 This is a schematic diagram of clustering experiment results according to another embodiment of this disclosure;

[0053] Figure 28 This is a schematic diagram of clustering experiment results according to another embodiment of this disclosure;

[0054] Figure 29 This is a schematic diagram of clustering experiment results according to yet another embodiment of this disclosure;

[0055] Figure 30 This is a schematic diagram of clustering experiment results according to yet another embodiment of this disclosure;

[0056] Figure 31 This is a schematic diagram of clustering experiment results according to yet another embodiment of this disclosure;

[0057] Figure 32 This is a schematic diagram of clustering experiment results according to yet another embodiment of this disclosure;

[0058] Figure 33 This is a schematic diagram of clustering experiment results according to yet another embodiment of this disclosure;

[0059] Figure 34 This is a schematic diagram of clustering experiment results according to yet another embodiment of this disclosure;

[0060] Figure 35 This is a schematic diagram showing the result of a high-density spatial metabolic data generation method according to an embodiment of the present disclosure.

[0061] Figure 36 This is a schematic diagram of the structure of a spatial metabolic data acquisition device according to an embodiment of the present disclosure. Detailed Implementation

[0062] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0063] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0064] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0065] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0066] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0067] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0068] The inventors noted that in spatial metabolomics, the most common forms of MSI technology include secondary ion mass spectrometry (SIMS), matrix-assisted laser desorption / ionization (MALDI), and desorption electrospray ionization (DESI). These MSI technologies offer unique advantages in terms of spatial resolution, detectable throughput, and tissue preservation.

[0069] Currently, both the commercial SCiLS Lab software package and the open-source tool cardinal cover basic MSI data preprocessing functions, including data import, standardization, smoothing, baseline correction, peak extraction, filtering, and alignment of metabolite annotations. pySM and metaspace provide methods for automated annotation of spatial metabolomics data.

[0070] In spatial transcriptomics, Scanpy and Seurat are popular tools. Their advantages lie in their large user community and rich documentation resources, and they support functions such as data filtering, dimensionality reduction, clustering, differential expression analysis, trajectory inference, and visualization.

[0071] MALDIpy and SpaMTP have embedded spatial metabolomics data analysis into Scanpy and Seurat, respectively, which not only expands the workflow of spatial metabolomics analysis but also makes multi-omics integrated analysis possible.

[0072] Currently, while models such as SpatialGlue, Nichecompass, and COSMOS have achieved the integration of spatial multi-omics modalities and revealed cell type distribution, intercellular interactions, and tissue functional zoning, the integration of spatial transcriptomics and spatial metabolomics is still limited by differences in spatial resolution.

[0073] With the rapid development of spatial multi-omics technologies, the SMA (Spatial Multimodal Analysis protocol) has enabled simultaneous and precise spatial transcriptomics processing and MALDI-MSI measurement on a single tissue slice, preserving the specificity and sensitivity of both modalities. However, the spatial coordinates of the spatial transcriptome and metabolome data generated by this method are still not in the same reference coordinate system. Therefore, precise point-to-point alignment of the spatial transcriptome and metabolome data remains a challenge during the integration process.

[0074] Furthermore, while MALDI-MSI provides higher resolution imaging information compared to DESI-MSI, MALDI-TOF (Matrix-Assisted Laser Desorption / Ionization Time-of-Flight Mass Spectrometry) often suffers from photodamage and photobleaching during high-resolution imaging. The number of metabolites detected decreases as the laser weakens, especially high mass-to-charge ratio metabolic ions. Therefore, improving the signal-to-noise ratio of spatial metabolic imaging information is also a challenge. Moreover, some fragile biological tissues or samples containing high water content are damaged during high-resolution imaging, problems that are less noticeable in low-resolution imaging. While the MOSR transfer learning framework can improve the resolution of spatial mass spectrometry imaging, it relies on a large number of high-quality optical images and supervised transfer learning using paired MSI data. Therefore, unsupervised methods for generating high-density spatial metabolic data are still lacking.

[0075] Accordingly, this disclosure provides a model training method that uses a generative adversarial network model to learn the correlation between the coordinates of each point and the spatial metabolic data of each point. This method can accurately predict the spatial metabolic data of the target point, so as to achieve precise point-to-point alignment of spatial transcriptome data and spatial metabolome data, and make the integration between spatial transcriptome and spatial metabolome not limited by spatial resolution differences.

[0076] In some embodiments, the model training method can also utilize low-resolution MSI data to generate high-resolution MSI data, thereby achieving unsupervised generation of high-density spatial metabolic data.

[0077] Figure 1 This is a schematic flowchart of a model training method according to an embodiment of the present disclosure. In some embodiments, the following model training method is performed by a model training device, including steps 101-110.

[0078] In step 101, mass spectrometry imaging is performed on the biological tissue slice sample to obtain a mass spectrometry image sample.

[0079] In step 102, the biological tissue section samples are stained with HE (hematoxylin-eosin) to obtain stained image samples.

[0080] In step 103, the mass spectrometry image sample and the stained image sample are registered to obtain the target image sample.

[0081] Figure 2 This is a schematic diagram illustrating the image registration process according to an embodiment of this disclosure. Figure 2As shown, mass spectrometry imaging was performed on biological tissue section sample 21 to obtain mass spectrometry image sample 22. HE staining was then performed on biological tissue section sample 21 to obtain stained image sample 23. Mass spectrometry image sample 22 and stained image sample 23 were registered to obtain target image sample 24.

[0082] It should be noted here that, as Figure 2 As shown, mass spectrometry image sample 22 can be represented by a series of ion intensity distribution maps with different mass-to-charge ratio (m / z) values. In other embodiments, the mass spectrometry image sample can also be represented by superimposing the ion intensity distributions of all mass-to-charge ratio values ​​onto the same image. Stained image sample 23 is a grayscale image that has been grayscaled.

[0083] In some embodiments, spatial metabolic data with highly spatially variable mass-to-charge ratios are extracted from mass spectrometry image samples to generate image samples to be processed. Using an affine transformation matrix, each pixel in the image samples to be processed is mapped to a stained image coordinate system to obtain the target image sample.

[0084] It should be noted that the mass spectrometry image sample includes spatial metabolic data of all mass-to-charge ratio values ​​obtained by mass spectrometry imaging. The mass-to-charge ratio value with high spatial variability is used as the target mass-to-charge ratio value. The spatial metabolic data of the target mass-to-charge ratio value is extracted from the mass spectrometry image sample to generate the image sample to be processed.

[0085] For example, mass spectrometry image samples are rasterized to construct an `anndata` object for mass spectrometry imaging data. The `spatial_autocorr` method in Squidpy is then used to extract spatial metabolic data of the target mass-to-charge ratio to highlight tissue contours and detailed tissue features. The extracted spatial metabolic data of the target mass-to-charge ratio are then summed and image normalization is performed to obtain the image sample to be processed. .

[0086] For example, if the mass-to-charge ratio satisfies ,as well as This indicates that the ion intensity of the mass-to-charge ratio is spatially concentrated and distributed, and the mass-to-charge ratio has high spatial variability. It is an index that measures spatial autocorrelation and is used to test whether the ion intensities of adjacent spatial locations are similar.

[0087] For example, stained image samples also undergo grayscale and normalization processing; the stained image samples are... .

[0088] In some embodiments, the original transformation matrix is ​​trained to obtain the affine transformation matrix.

[0089] In some embodiments, the method for training the original transformation matrix includes the following steps S11-S15.

[0090] In step S11, the original transformation matrix is ​​generated based on the horizontal scaling factor, vertical scaling factor, horizontal shearing factor, vertical shearing factor, horizontal translation amount, and vertical translation amount.

[0091] For example, suppose the horizontal scaling factor is Vertical scaling factor is The transverse shear factor is Longitudinal shear factor is The lateral translation amount is And longitudinal translation amount is Then the original transformation matrix As shown in formula (1).

[0092] (1)

[0093] In step S12, the first coordinates of each pixel in the image sample to be processed are transformed into the stained image sample using the original transformation matrix to obtain the second coordinates of each pixel.

[0094] In step S13, the coordinates of feature points in the stained image sample that have a one-to-one correspondence with each pixel are extracted.

[0095] In step S14, the mean square error of the coordinates of the second coordinates of each pixel and the coordinates of the feature points corresponding to each pixel is calculated.

[0096] In step S15, based on the mean square error of the coordinates, the horizontal scaling factor, vertical scaling factor, horizontal shearing factor, vertical shearing factor, horizontal translation amount, and vertical translation amount are trained to obtain the affine transformation matrix.

[0097] It's important to note that the affine transformation matrix can be calculated by minimizing the mean square error of the coordinates. In other words, the affine transformation matrix is ​​used to transform each pixel in the image sample to be as close as possible to the coordinates of the feature points in the stained image sample that have a one-to-one correspondence with each pixel.

[0098] For example, suppose the image sample to be processed The total number of pixels in the image is P, and the first coordinate of the p-th pixel is... , The original transformation matrix is stained image samples The coordinates of the feature point that has a one-to-one correspondence with the p-th pixel are: Then the mean square error of the coordinates As shown in formula (2).

[0099] (2)

[0100] For example, by minimizing the mean square error of the coordinates The affine transformation matrix is ​​calculated as follows: .

[0101] It should be noted that, by using an affine transformation matrix, each pixel in the image sample to be processed is mapped to the stained image coordinate system, and the pixel values ​​in the image sample to be processed are redistributed to the stained image coordinate system to obtain the target image sample.

[0102] For example, the image sample to be processed The coordinates of the p-th pixel in the image are: Using affine transformation matrix Transform the p-th pixel to obtain the p-th pixel in the target image sample. coordinates in As shown in formula (3).

[0103] (3)

[0104] For example, the image sample to be processed In coordinates The pixel value at that location is Then the target image sample In coordinates pixel value at As shown in formula (4).

[0105] (4)

[0106] In step 104, spatial metabolic data of each point in a specified point set in the target image sample is obtained.

[0107] For example, a specified set of points in a target image sample includes points obtained by transforming P pixels in the image sample to be processed using an affine transformation matrix.

[0108] In some embodiments, the spatial metabolic data of the i-th point is a data vector comprising M elements, wherein the j-th element of the data vector represents the metabolic intensity of the i-th point at the j-th mass-to-charge ratio. N is the total number of points in the specified set. M represents the total number of mass-to-charge ratios.

[0109] It should be noted that metabolic intensity can also be called ionic intensity.

[0110] For example, the spatial metabolic data of the i-th point As shown in formula (5).

[0111] (5)

[0112] For example, high-density spatial metabolic data can be generated using the spatial metabolic data of each point in a specified point set to improve the resolution of the spatial metabolic data; and the generated high-density spatial metabolic data can be used to train the model, thereby improving the training effect of the model.

[0113] In some embodiments, the high-density spatial metabolic data generation method includes the following steps S21-S23.

[0114] In step S21, multiple points are selected within the neighborhood of each point in the specified point set to add.

[0115] In some embodiments, a polar coordinate system is established for each point, where each point is the pole of the polar coordinate system, and the ray extending from the pole along a predetermined direction is the polar axis of the polar coordinate system. Corresponding points of multiple candidate coordinates in the polar coordinate system are used as multiple additional points.

[0116] In some embodiments, the plurality of candidate coordinates includes a first candidate coordinate. Second candidate coordinates and the third candidate coordinates , where R is the predetermined distance.

[0117] Figure 3 This is a schematic diagram of a high-density spatial metabolic data generation method according to an embodiment of this disclosure. Figure 3 As shown, the specified point set includes point 31, point 32, point 33, and point 34.

[0118] As an example, a polar coordinate system is established for point 31, where point 31 is the pole of the system, and ray 35 extending from point 31 along a predetermined direction is the polar axis of the system. The coordinates in this polar coordinate system... According to the polar diameter and polar angle Sure.

[0119] Figure 4 This is a schematic diagram of a high-density spatial metabolic data generation method according to another embodiment of this disclosure. Figure 4 As shown, the multiple points selected in the neighborhood of point 31 include point 311, point 312, and point 313.

[0120] As an example, in a polar coordinate system with point 31 as the pole and ray 35 as the polar axis, adding point 311 would be the first candidate coordinate in this polar coordinate system. The corresponding point, point 312, is the second candidate coordinate in this polar coordinate system. The corresponding point, point 313, is the third candidate coordinate in this polar coordinate system. The corresponding point.

[0121] For example, if the distance between point 31 and point 32 is Then R can be set to .

[0122] Figure 5 This is a schematic diagram of a high-density spatial metabolic data generation method according to yet another embodiment of this disclosure. Figure 5 As shown, the multiple points selected in the neighborhood of point 31 include points 311, 312, and 313; the multiple points selected in the neighborhood of point 32 include points 321, 322, and 323; the multiple points selected in the neighborhood of point 33 include points 331, 332, and 333; and the multiple points selected in the neighborhood of point 34 include points 341, 342, and 343.

[0123] It should be noted that polar coordinate systems are established for points 32, 33, and 34 respectively, and the corresponding points of multiple candidate coordinates in the polar coordinate system corresponding to each point are used as the added points for each point.

[0124] In step S22, the spatial metabolic data of the points in the specified set of points closest to the addition point are used as the spatial metabolic data of the addition point.

[0125] For example, for point 311, point 31 in the specified point set is closest to point 311. Therefore, the spatial metabolic data of point 31 is used as the spatial metabolic data of point 311.

[0126] In step S23, the spatial metabolic data of the kth addition point is updated using the spatial metabolic data of the 8 neighboring points of the kth addition point. K is the total number of points to be added to the specified point set.

[0127] For example, such as Figure 5 As shown, the eight neighboring points of point 311 include point 313, point 312, point 323, point 31, point 32, point 333, point 332, and point 343. The spatial metabolic data of point 311 is updated using the spatial metabolic data of the eight neighboring points of point 311.

[0128] It should be noted that for a point added at the edge, if it has fewer than 8 neighboring points, the point can be updated using only all its neighboring points. Alternatively, a ring of expanding points can be generated outwards, so that the point added at the edge has 8 neighboring points. The spatial metabolic data of the expanding point is then used from the spatial metabolic data of the point added closest to the expanding point or from a specified set of points.

[0129] For example, the neighboring points of point 313 include point 312, point 311, and point 31. Point 313 can update the spatial metabolic data using only the above three neighboring points.

[0130] For example, a link prediction model can be used to update the spatial metabolic data of added points, achieving self-supervised generation of high-density spatial metabolic data. The link prediction model includes an STMGraph model, which uses graph attention for encoding and bilateral decoding, with random masks applied before both encoding and decoding. The topology graph selects 8 nearest neighbors and constructs an adjacency matrix with initial edge weights of 1. Specific parameters for training the STMGraph model include num_epoch, lr, weight_decay, hidden_dims, mask_ratio, and noise, where num_epoch is set to 1000, lr to 0.001, weight_decay to 1e-4, hidden_dims to [256, 30], mask_ratio to 0.5, and noise to 0.0.

[0131] For example, the resolution of spatial metabolic data for a specified point set is The resolution of the high-density spatial metabolic data generated using the method described in the above embodiments is [missing information]. This improves the resolution of spatial metabolic data.

[0132] The method described in the above embodiments generates high-density spatial metabolic data using spatial metabolic data of each point in a specified point set, thereby improving the resolution of the spatial metabolic data; and the generated high-density spatial metabolic data is used to train the model, thereby improving the training effect of the model.

[0133] In step 105, the encoder in the generative adversarial network model is used to process the spatial metabolic data of each point to obtain the predicted coordinates of each point.

[0134] It should be noted that the generative adversarial network model consists of an autoencoder and a determiner. The autoencoder includes an encoder and a decoder, wherein the encoder in the autoencoder is used as the generator in this disclosure.

[0135] Figure 6 This is a schematic diagram of the structure of a generative adversarial network model according to an embodiment of this disclosure. Figure 6 As shown, the generative adversarial network model 60 includes an encoder 61, a decoder 62, and a decision 63.

[0136] For example, using Figure 6The encoder 61 in the generative adversarial network model 60 shown processes the spatial metabolic data of each point to obtain the predicted coordinates of each point.

[0137] For example, suppose the spatial metabolic data of the i-th point is The encoder in the generative adversarial network model is Then the predicted coordinates of the i-th point As shown in formula (6).

[0138] (6)

[0139] Among them, the spatial metabolic data of the i-th point The size is .

[0140] In step 106, the predicted coordinates of each point are processed using the decoder in the generative adversarial network model to obtain the spatial metabolic reconstruction data of each point.

[0141] For example, using Figure 6 The decoder 62 in the generative adversarial network model 60 shown processes the predicted coordinates of each point to obtain the spatial metabolic reconstruction data of each point.

[0142] For example, suppose the predicted coordinates of the i-th point are The decoder in the generative adversarial network model is Then the spatial metabolic reconstruction data of the i-th point As shown in formula (7).

[0143] (7)

[0144] In step 107, the discriminator in the generative adversarial network model is used to identify the probability that the coordinates of each point are true coordinates, which is used as the confidence level of the coordinates of each point.

[0145] For example, using Figure 6 The discriminator 63 in the generative adversarial network model 60 shown identifies the probability that the coordinates of each point are true coordinates, and uses this as the confidence level of the coordinates of each point.

[0146] For example, suppose the coordinates of the i-th point are In a generative adversarial network model, if the discriminator is G, then the confidence level of the coordinates of each point is: .

[0147] In step 108, the discriminator is used to identify the probability that the predicted coordinates of each point are the true coordinates, which is used as the confidence level of the predicted coordinates of each point.

[0148] For example, using Figure 6The discriminator 63 shown identifies the probability that the predicted coordinates of each point are the true coordinates, and uses this as the confidence level of the predicted coordinates of each point.

[0149] For example, suppose the predicted coordinates of the i-th point are If the discriminator in the generative adversarial network model is G, then the confidence level of the predicted coordinates of each point is: .

[0150] In step 109, the loss value is determined based on the spatial metabolic data of each point, the spatial metabolic reconstruction data of each point, the coordinates of each point, the predicted coordinates of each point, the confidence level of the coordinates of each point, and the confidence level of the predicted coordinates of each point.

[0151] In some embodiments, the method for determining the loss value includes the following steps S31-S33.

[0152] In step S31, the first sub-loss value is determined based on the spatial metabolic data and the spatial metabolic reconstruction data of each point.

[0153] In some embodiments, the weighted mean square error of the spatial metabolic data and the spatial metabolic reconstruction data at each point is calculated to obtain a first bias value. The mean absolute error of the spatial metabolic data and the spatial metabolic reconstruction data at each point is calculated to obtain a second bias value. The weighted sum of the first bias value and the second bias value is calculated to obtain a first sub-loss value.

[0154] It should be noted here that the weighted sum of the first deviation value and the second deviation value is determined based on the weight of the first deviation value and the weight of the second deviation value.

[0155] In some embodiments, the weight of the first deviation value is determined.

[0156] In some embodiments, the method for determining the weight of the first deviation value includes: constructing a metabolic intensity matrix, wherein the metabolic intensity matrix includes spatial metabolic data of N points in a specified point set; calculating the sum of all metabolic intensities for M mass-to-charge ratios in the metabolic intensity matrix to obtain a matrix sum; calculating the sum of all metabolic intensities for the j-th mass-to-charge ratio in the metabolic intensity matrix to obtain the j-th sum; calculating the ratio of the j-th sum to the matrix sum to obtain the j-th ratio; and calculating the difference between a predetermined value and the j-th ratio to obtain the j-th weight value in the weight of the first deviation value.

[0157] It should be noted that the metabolic intensity matrix can also be called the ionic intensity matrix, and the sum of all metabolic intensities can also be called the sum of all ionic intensities.

[0158] For example, the spatial metabolic data of the i-th point is Then the metabolic intensity matrix is Metabolic intensity matrix Includes spatial metabolic data for N points in a specified point set. Metabolic intensity matrix. The size is The element in the i-th row and j-th column Let be the ion intensity of the j-th mass-to-charge ratio at the i-th point. The weight of the first deviation value is . The size is The j-th element is the j-th weight value. .

[0159] For example, suppose the metabolic intensity matrix is The element in the i-th row and j-th column Let the ion intensity be the mass-to-charge ratio at point i, then the weight of the first deviation value is... The j-th weight value in As shown in formula (8).

[0160] (8)

[0161] The predetermined value is 1.

[0162] It should be noted here that, to prevent the model from tending to reconstruct features of mass-to-charge ratios with high ion strength, a weighted approach is used to calculate the first sub-loss value, assigning the j-th mass-to-charge ratio a weight value. .

[0163] For example, suppose the spatial metabolic data of the i-th point is The spatial metabolic reconstruction data of the i-th point is The weight of the first deviation value is Then the first sub-loss value As shown in formula (9).

[0164] (9)

[0165] in, The size is ; express The vector is squared element by element, and its size is also... ; Size is ,therefore express row vectors and The product of the column vectors yields a numerical value; The weight of the second deviation value.

[0166] For example, to make it easier to understand, the first sub-loss value It can also be written in the form of formula (10).

[0167] (10)

[0168] In step S32, the second sub-loss value is determined based on the coordinates of each point, the predicted coordinates of each point, the confidence level of the coordinates of each point, and the confidence level of the predicted coordinates of each point.

[0169] In some embodiments, the mean absolute error of the coordinates and predicted coordinates of each point is calculated to obtain a coordinate error value. The average confidence level of the coordinates of each point is calculated to obtain a first confidence mean. The average confidence level of the predicted coordinates of each point is calculated to obtain a second confidence mean. The weighted sum of the coordinate error value and the first confidence mean is calculated to obtain an intermediate loss value. The difference between the intermediate loss value and the second confidence mean is calculated to obtain a second sub-loss value.

[0170] For example, suppose the coordinates of the i-th point are The predicted coordinates of the i-th point are If the discriminant in the generative adversarial network model is G, then the second sub-loss value is... As shown in formula (11).

[0171] (11)

[0172] in, The weights are the coordinate error values.

[0173] In step S33, the loss value is obtained based on the first sub-loss value and the second sub-loss value.

[0174] In some embodiments, a weighted sum of the first sub-loss value and the second sub-loss value is calculated to obtain the loss value.

[0175] For example, suppose the first sub-loss value is The second sub-loss value is Then the loss value As shown in formula (12).

[0176] (12)

[0177] in, The weights of the first sub-loss values, The weights for the second sub-loss values.

[0178] In step 110, the encoder, decoder, and discriminator are trained using the loss values.

[0179] By using the model training method described in the above embodiments, the generative adversarial network model is used to learn the correlation between the coordinates of each point and the spatial metabolic data of each point. The encoder in the resulting generative adversarial network model can accurately predict the coordinates corresponding to the spatial metabolic data, and the decoder can accurately predict the spatial metabolic data of the target point. This enables point-to-point precise alignment of spatial transcriptome data and spatial metabolome data, so that the integration between spatial transcriptome and spatial metabolome is not limited by spatial resolution differences.

[0180] Figure 7 This is a schematic diagram of the structure of a model training apparatus according to an embodiment of this disclosure. Figure 7 As shown, the model training device 70 can be represented in the form of a general computing device. The model training device 70 includes a memory 71, a processor 72, and a bus 73 connecting different system components.

[0181] The memory 71 may include, for example, system memory, non-volatile storage media, etc. System memory may store, for example, an operating system, application programs, a boot loader, and other programs. System memory may include volatile storage media, such as random access memory (RAM) and / or cache memory. Non-volatile storage media may store, for example, instructions for a corresponding embodiment of an executing model training method. Non-volatile storage media include, but are not limited to, disk storage, optical storage, flash memory, etc.

[0182] The processor 72 can be implemented using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete hardware components such as discrete gates or transistors. Accordingly, each module, such as the acquisition module, the calculation module, and the adjustment module, can be implemented by executing instructions in the central processing unit (CPU) running memory to perform the corresponding steps, or by implementing dedicated circuits that perform the corresponding steps.

[0183] For example, processor 72 is configured to execute instructions stored in memory 71, such as Figure 1 The method involved in any of the embodiments.

[0184] Bus 73 can use any of the various bus architectures. For example, bus architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MCA) bus, and the Peripheral Component Interconnect (PCI) bus.

[0185] The interfaces 74, 75, and 76 of the model training device 70, as well as the memory 71 and processor 72, can be connected via bus 73. Input / output interface 74 provides a connection interface for input / output devices such as a monitor, mouse, and keyboard. Network interface 75 provides a connection interface for various networked devices. Storage interface 76 provides a connection interface for external storage devices such as floppy disks, USB flash drives, and SD cards.

[0186] The generative adversarial network model is trained using the model training device and model training method described in the above embodiments. The encoder in the resulting generative adversarial network model can accurately predict the coordinates corresponding to spatial metabolic data, and the decoder can accurately predict the spatial metabolic data of the target point. This enables point-to-point precise alignment of spatial transcriptome data and spatial metabolome data, allowing the integration between spatial transcriptome and spatial metabolome to be unrestricted by spatial resolution differences.

[0187] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations thereof, can be implemented by computer-readable program instructions.

[0188] These computer-readable program instructions are provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable device to produce a machine, such that execution of the instructions by the processor produces means for implementing the functions specified in one or more boxes of the flowchart and / or block diagram.

[0189] These computer-readable program instructions may also be stored in a computer-readable storage medium. These instructions cause a computer to work in a particular manner to produce an article of manufacture, including instructions that implement the functions specified in one or more boxes in a flowchart and / or block diagram.

[0190] This disclosure may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0191] This disclosure also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement... Figure 1 The method involved in any of the embodiments.

[0192] This disclosure also provides a computer program product, including computer instructions, wherein the computer instructions, when executed by a processor, implement as follows: Figure 1 The method involved in any of the embodiments.

[0193] Figure 8This is a schematic flowchart of a spatial metabolic data acquisition method according to an embodiment of the present disclosure. In some embodiments, the following spatial metabolic data acquisition method is performed by a spatial metabolic data acquisition device, including steps 81-85.

[0194] In step 81, mass spectrometry imaging is performed on the biological tissue slices to be processed to obtain mass spectrometry images.

[0195] In step 82, the biological tissue sections to be processed are stained with hematoxylin and eosin (HE) to obtain stained images.

[0196] In step 83, the mass spectrometry image and the staining image are registered to obtain the target image.

[0197] In some embodiments, spatial metabolic data with highly spatially variable mass-to-charge ratios are extracted from mass spectrometry images to generate an image to be processed. Using an affine transformation matrix, each pixel in the image to be processed is mapped to a stained image coordinate system to obtain the target image.

[0198] In step 84, the coordinates of each measurement point among multiple measurement points in the target image are obtained.

[0199] In step 85, the coordinates of each measurement point are processed using a decoder to obtain the spatial metabolic data for each measurement point. The decoder is trained using the model training method described in any of the above embodiments.

[0200] The spatial metabolic data acquisition method described in the above embodiments uses the decoder in the trained generative adversarial network model to process the coordinates of each measurement point, thereby obtaining the spatial metabolic data of each measurement point. This enables precise point-to-point alignment of spatial transcriptome data and spatial metabolome data, ensuring that the integration between spatial transcriptome and spatial metabolome is not limited by spatial resolution differences.

[0201] In some embodiments, mass spectrometry imaging is performed on the biological tissue sections to be processed to obtain mass spectrometry images. HE staining is performed on the biological tissue sections to be processed to obtain stained images. The mass spectrometry images and stained images are registered to obtain a target image. Spatial transcriptomics processing is performed on the biological tissue sections to be processed to obtain spatial transcription data for each transcription sampling point in the transcription sampling point set. The coordinates of each measurement point in a plurality of measurement points in the target image are obtained, wherein the plurality of measurement points are transcription sampling points in the transcription sampling point set. The coordinates of each measurement point are processed using a decoder to obtain spatial metabolic data for each measurement point, so as to achieve precise point-to-point alignment of the spatial metabolic data and spatial transcription data for each measurement point, wherein the decoder is trained using the model training method described in any of the above embodiments.

[0202] It should be noted that, on the one hand, spatial metabolic data of each metabolic sampling point in the spatial metabolic sampling point set is obtained through mass spectrometry imaging; on the other hand, the coordinates of each metabolic sampling point are transformed into the coordinate system of the stained image through registration of the mass spectrometry image and the stained image, corresponding to a specified set of points in the target image; the decoder is trained using the coordinates of each point in the specified set of points and the spatial metabolic data of each point, so that the decoder can learn the relationship between the coordinates and the spatial metabolic data.

[0203] It should also be noted that, on the other hand, spatial transcriptomics processing is used to obtain the spatial transcription data of each transcription sampling point in the spatial transcription sampling point set. Since the spatial transcription sampling point set and the HE staining image are in the same coordinate system, the coordinates of multiple measurement points corresponding to the spatial transcription sampling point set in the target image can be directly obtained. The spatial transcription data of each measurement point is the spatial transcription data of the corresponding transcription sampling point. The coordinates of each measurement point are processed using a trained decoder to obtain the spatial metabolic data of each measurement point.

[0204] The method described in the above embodiments obtains spatial metabolic data and spatial transcription data for each measurement point among multiple measurement points in the target image, achieving precise point-to-point alignment of spatial transcriptome data and spatial metabolome data, thereby realizing the integration of spatial metabolome data and spatial transcriptome data.

[0205] Figure 9 This is a schematic flowchart illustrating a spatial metabolic data acquisition method according to another embodiment of this disclosure.

[0206] like Figure 9 As shown, mass spectrometry imaging was performed on the biological tissue sections to be processed to obtain spatial metabolomics data. Figure 91 shows the distribution of sampling points for the spatial metabolomics data. For example, the resolution of the spatial metabolomics sampling point set is... .

[0207] like Figure 9 As shown, spatial transcriptomics processing was performed on the biological tissue sections to be processed, yielding spatial transcriptomics data. Figure 92 shows the distribution of sampling points for the spatial transcriptomics data. For example, the resolution of the spatial transcriptomics sampling point set is... .

[0208] It should be noted that, as shown in sampling point distribution diagrams 91 and 92, the distribution of sampling points for spatial metabolomics data and spatial transcriptomics data may differ.

[0209] like Figure 9As shown, using the method described in the above embodiments, the coordinates of each measurement point in the target image are processed using a decoder to obtain the spatial metabolic data of each measurement point. Each measurement point is a sampling point for spatial transcriptome data. The spatial metabolic data of multiple measurement points in the target image are visualized to obtain ion intensity distribution map 93. Ion intensity distribution map 93 shows the ion intensity distribution of multiple measurement points in the target image.

[0210] like Figure 9 As shown, clustering was performed using spatial metabolic and spatial transcriptional data from multiple measurement points in the target image, resulting in clustering result 94.

[0211] The methods described in the above embodiments enable precise point-to-point alignment of spatial transcriptomic data and spatial metabolomic data.

[0212] Figure 10 This is a schematic diagram of a high-density spatial metabolic data generation method according to yet another embodiment of this disclosure. Figure 10 As shown, the resolution of the original spatial metabolic data 101 is... The original spatial metabolic data 101 is processed using the high-density spatial metabolic data generation method described in the above embodiments to obtain high-density spatial metabolic data 102, wherein the resolution of the high-density spatial metabolic data 102 is [resolution missing]. .

[0213] Figure 11 This is a schematic diagram of a high-density spatial metabolic data generation method according to yet another embodiment of this disclosure. Figure 11 As shown, the raw spatial metabolic data 111 was obtained by mass spectrometry imaging of mouse brain tissue slices, with a resolution of [resolution missing]. The high-density spatial metabolic data 112 generated from the original spatial metabolic data 111 has a resolution of [missing information]. .

[0214] The methods described in the above embodiments use HE-stained images as a medium to automatically align MSI and HE in the same reference coordinate system through end-to-end training of an affine transformation matrix. An adversarial autoencoder is constructed to associate spatial coordinates with spatial metabolites, and finally, using spatial transcriptome coordinate information, the corresponding spatial metabolic expression products are predicted. Furthermore, the self-supervised training model STMGraph, which has link prediction capabilities, can be used to generate high-density mass spectrometry imaging data, resulting in better preservation of high-quality metabolic ion information under high-resolution imaging.

[0215] It should be noted that the generative adversarial network model used in the above training method is SM2ST, which supports mainstream operating systems (such as Windows and Linux). Users need to select the appropriate installation package according to their own operating system. The required graphics card is an NVIDIA GeForce RTX 4080 16G, with at least 32GB of RAM and a 64-core processor. SM2ST supports various common mass spectrometry data formats (such as .mzML and csv). The software requires the cardinal data preprocessing module to run. Software dependencies include: python (version 3.9.19), numpy (version 1.26.4), squidpy (version 1.6.1), scanpy (version 1.9.8), r-base (version 4.2.2), rpy2 (version 3.5.9), torch-cluster (version 1.6.1+pt113cu117), torch-geometric (version 2.5.3), torch-scatter (version 2.1.1+pt113cu117), torch-sparse (version 0.6.17+pt113cu117), torch-spline-conv (version 1.2.2+pt113cu117), pytorch (version 1.13.1), pytorch-cuda (version 11.7), SpatialGlue, and STMGraph.

[0216] The method for acquiring spatial metabolic data will be described below through specific embodiments.

[0217] In some embodiments, the method for preparing biological tissue sections includes the following steps S41-S44.

[0218] In step S41, the complete brain tissue of the mouse is removed and rapidly frozen in liquid nitrogen to maximize the stability of the tissue's endogenous metabolites.

[0219] In step S42, at In a cryostat, multiple coronal sections were prepared using intact mouse brain tissue, with each coronal section having a thickness of [missing information]. .

[0220] In step S43, multiple coronal slices are mounted on a conductive glass slide with an indium tin oxide (ITO) coating and stored in... In ultra-low temperature freezers.

[0221] In step S44, mass spectrometry imaging is performed on multiple coronal sections to obtain two mass spectrometry imaging datasets of the left and right brain regions of the mouse, with a spatial resolution of [missing information]. And a mass spectrometry imaging dataset of adjacent slices of the right brain, with a spatial resolution of .

[0222] It should be noted that after the mass spectrometry imaging process is completed, all acquired pixel data are exported uniformly in the standard IMZML format. ShinyCardinal is used for data standardization, including: peak picking using a MAD (mean absolute deviation) noise filter with a MAD ratio of 4, peak alignment with a peak alignment of 15 ppm, normalization using root mean square, smoothing using the Savitzky-Golay algorithm, baseline elimination using local minima, and peak binning with a tolerance of 15 ppm.

[0223] For example, according to the SMA protocol, spatial transcriptomic and spatial metabolomic data were obtained from the brains of three mice: V11L12-038, V11T16-085, and V11L12-109, with three samples from each mouse. V11L12-038 received one sample using 9-AA as the matrix for metabolite detection (negative mode) and two samples using DHB for metabolite detection (positive mode), one of which served as a control group. V11T16-085 received three samples using FMP-10 as the matrix for metabolite detection (positive mode). Similarly, V11L12-109 received three samples using FMP-10 as the matrix (positive mode). The multiple samples obtained from these three mice are shown in Table 1.

[0224] Table 1

[0225] Figure 12 This is a flowchart illustrating a spatial metabolic data acquisition method according to yet another embodiment of this disclosure.

[0226] like Figure 12 As shown, mass spectrometry imaging was performed on mouse brain tissue slices to obtain mass spectrometry image 121, which includes spatial metabolic data of all mass-to-charge ratios. Spatial metabolic data of mass-to-charge ratios with high spatial variability were extracted from the mass spectrometry image to obtain image 122 to be processed. HE staining was performed on mouse brain tissue slices to obtain staining image 123, which is a grayscale image after grayscale conversion.

[0227] like Figure 12As shown, the image to be processed 122 and the stained image 123 are registered to obtain the target image, which includes spatial metabolic data with multiple mass-to-charge ratios. As an example, ion intensity distribution map 124 shows the ion intensity distribution of the target image with a mass-to-charge ratio of 156.26. The high-intensity signal at the corpus callosum (cc) (as shown by boxes Q1 and Q2 in ion intensity distribution map 124) is shown in corpus callosum detail map 125 and corpus callosum detail map 126.

[0228] like Figure 12 As shown, the decoder in the generative adversarial network model processes each of the multiple measurement points in the target image to obtain the spatial metabolic data for each measurement point. This spatial metabolic data from multiple measurement points in the target image is then used as metabolic prediction data. As an example, Figure 127 shows the ion intensity distribution for a mass-to-charge ratio of 156.26 in the metabolic prediction data.

[0229] It should be noted that since staining images 123 have already been registered and aligned with the spatial transcriptome data, in order to integrate the spatial metabolome data and spatial transcriptome data at the same spot, staining image 123 is first used as a bridge to align the spatial metabolome data and staining image 123 in the same spatial coordinate system. Furthermore, directly superimposing all mass-to-charge ratio spatial metabolic data (as shown in mass spectrometry image 121) will not highlight the outline and tissue details of the mouse brain; if... Using a threshold greater than 0.5 as a selection criterion, spatial metabolic data with highly variable mass-to-charge ratios are extracted and superimposed (as shown in image 122). This allows for the visualization of the true contours of mouse brain slices and details of the corpus callosum. Through an affine transformation matrix, the mouse brain contours and corpus callosum in image 122 and stained image 123 can be registered one-to-one, as shown in ion intensity distribution map 124.

[0230] It is also worth noting that by remapping the spatial metabolic data to the coordinates of each site in the spatial transcriptome through the decoder in the generative adversarial network model, the metabolite with a mass-to-charge ratio of 156.26 not only retains low-intensity information in neurons projecting to the dorsal putamen of the striatum in the substantia nigra pars compacta (SNc), but also clearly restores the high-intensity signal in the corpus callosum, as shown in detail figures 125 and 126 of the corpus callosum. Since the spatial metabolic data for each measurement point takes spatial location information into account, the signal at each measurement point is influenced by the signal in the local region, which can enhance the spatial specificity of the ion signal.

[0231] In some embodiments, a resolution of Benchmark experiments were conducted using MSI data from the left and right hemispheres of mice to compare the performance of the STAGE model with that of the SM2ST generative adversarial network model trained using the model training methods described in the above embodiments. The STAGE model is a super-resolution model for spatial transcriptomics, using spatial coordinates to predict transcriptomic expression at unmeasured sites.

[0232] For example, construct anndata objects for MSI data and randomly dropout 50% of the points to compare the spatial metabolic data restoration capabilities of the SM2ST model and the STAGE model.

[0233] It should be noted that MSI data can also be called spatial metabolic data.

[0234] For example, 50% of the points are randomly masked from the original MSI data to obtain missing data. The STAGE model is used to process the missing data to generate the first restored MSI data. The SM2ST model is then used to process the missing data to generate the second restored MSI data.

[0235] Figure 13 This is a schematic diagram of the benchmark test results of one embodiment of the present disclosure. Figure 13 The results are from a benchmark test with a mass-to-charge ratio of 152.06. Figure 13 As shown, ion intensity distribution diagram 131 shows the ion intensity distribution of a mass-to-charge ratio of 152.06 in the original MSI data. Ion intensity distribution diagram 132 shows the ion intensity distribution of a mass-to-charge ratio of 152.06 in the missing data. Ion intensity distribution diagram 133 shows the ion intensity distribution of a mass-to-charge ratio of 152.06 in the first reduced MSI data. Ion intensity distribution diagram 134 shows the ion intensity distribution of a mass-to-charge ratio of 152.06 in the second reduced MSI data.

[0236] Figure 14 This is a schematic diagram of benchmark test results for another embodiment of this disclosure. Figure 14 The results are from a benchmark test with a mass-to-charge ratio of 159.00. Figure 14 As shown, ion intensity distribution diagram 141 shows the ion intensity distribution for a mass-to-charge ratio of 159.00 in the original MSI data. Ion intensity distribution diagram 142 shows the ion intensity distribution for a mass-to-charge ratio of 159.00 in the missing data. Ion intensity distribution diagram 143 shows the ion intensity distribution for a mass-to-charge ratio of 159.00 in the first reduced MSI data. Ion intensity distribution diagram 144 shows the ion intensity distribution for a mass-to-charge ratio of 159.00 in the second reduced MSI data.

[0237] like Figure 14As shown, box Q3 in ion intensity distribution map 141, box Q4 in ion intensity distribution map 143, and box Q5 in ion intensity distribution map 144 are all regions of the corpus callosum.

[0238] Figure 15 This is a schematic diagram of the benchmark test results for yet another embodiment of this disclosure. Figure 15 The results are from a benchmark test with a mass-to-charge ratio of 152.01. Figure 15 As shown, ion intensity distribution diagram 151 shows the ion intensity distribution of a mass-to-charge ratio of 152.01 in the original MSI data. Ion intensity distribution diagram 152 shows the ion intensity distribution of a mass-to-charge ratio of 152.01 in the missing data. Ion intensity distribution diagram 153 shows the ion intensity distribution of a mass-to-charge ratio of 152.01 in the first reduced MSI data. Ion intensity distribution diagram 154 shows the ion intensity distribution of a mass-to-charge ratio of 152.01 in the second reduced MSI data.

[0239] Figure 16 This is a schematic diagram of the benchmark test results for yet another embodiment of this disclosure. Figure 16 The results are from a benchmark test with a mass-to-charge ratio of 228.04. Figure 16 As shown, ion intensity distribution diagram 161 shows the ion intensity distribution of the original MSI data with a mass-to-charge ratio of 228.04. Ion intensity distribution diagram 162 shows the ion intensity distribution of the missing data with a mass-to-charge ratio of 228.04. Ion intensity distribution diagram 163 shows the ion intensity distribution of the first reduced MSI data with a mass-to-charge ratio of 228.04. Ion intensity distribution diagram 164 shows the ion intensity distribution of the second reduced MSI data with a mass-to-charge ratio of 228.04.

[0240] Figure 17 This is a schematic diagram of benchmark test results for yet another embodiment of this disclosure. Figure 17 As shown, the horizontal axis represents the specific model, including the STAGE model and the SM2ST model; the vertical axis represents the Pearson correlation. Figure 17 The Pearson correlation of the first restored MSI data generated by the STAGE model and the Pearson correlation of the second restored MSI data generated by the SM2ST model are shown, along with their confidence levels. .

[0241] Figure 18 This is a schematic diagram of benchmark test results for yet another embodiment of this disclosure. Figure 18 As shown, the horizontal axis represents the specific model, including the STAGE model and the SM2ST model; the vertical axis represents the Pearson correlation. Figure 18This paper presents the Pearson correlation of the first restored MSI data generated by the STAGE model and the Pearson correlation of the second restored MSI data generated by the SM2ST model in another sample, along with their confidence levels. .

[0242] It should be noted here that, as Figure 13 , Figure 15 and Figure 16 As shown, the STAGE model cannot reproduce low-intensity metabolic information, but the SM2ST model can reproduce it very well. Figure 14 As shown, the data reconstruction results of the STAGE model can only present the information for the mass-to-charge ratio of 159.00. The signal in box Q4 is inconsistent with the signal in box Q3, indicating that the STAGE model cannot correctly reconstruct the corpus callosum information. Conversely, the signal in box Q5 is consistent with the signal in box Q3, indicating that the SM2ST model can accurately reconstruct the corpus callosum information and smooth the fracture signal.

[0243] It should also be noted here that, as Figure 17 and Figure 18 As shown, the data generated by the SM2ST model and the STAGE model are significantly different. The median Pearson correlations for the data reconstructed by the SM2ST model were 0.968 and 0.971, respectively, while those for the data reconstructed by the STAGE model were 0.967 and 0.961. This indicates that the SM2ST model exhibits higher stability and consistency in the data reconstruction process, with a higher median Pearson correlation and the data being more concentrated around the median. Although the STAGE model can also achieve high correlations in some cases, the overall data is more dispersed, and some data points have lower correlations, which are precisely the low-intensity metabolic signals.

[0244] It's also worth noting that the SM2ST model's performance benefits from its adversarial architecture and weighted loss function. By introducing an adversarial training mechanism, the SM2ST model generates more continuous coordinate samples, making the latent space smoother and more continuous, facilitating the perception of metabolic intensity at neighboring points in unknown locations. Simultaneously, the weighted loss function balances high-intensity and low-intensity ion signals, allowing the network to learn richer information about changes.

[0245] In some embodiments, spatial metabolomics and spatial transcriptomics data from mouse brain slices are used to perform multi-omics data integration experiments to compare the retention of metabolic signals with spatially highly variable mass-to-charge ratios in the generated data of the STAGE model and the SM2ST model, in order to demonstrate the high variability of metabolic signal retention of the SM2ST model.

[0246] It should be noted that metabolic signals with spatially highly variable mass-to-charge ratios can also be called highly variable metabolic signals, and metabolic ions corresponding to spatially highly variable mass-to-charge ratios can also be called highly variable metabolic ions.

[0247] For example, based on the SMA protocol, spatial transcriptomic and spatial metabolomic data of the same mouse brain slice are acquired. The STAGE model is used to integrate the spatial transcriptomic and spatial metabolomic data of the same mouse brain slice through multi-omics data fusion, generating spatial metabolic data corresponding to the coordinates of the spatial transcriptomic data, resulting in the first derived data. The SM2ST model is then used to integrate the spatial transcriptomic and spatial metabolomic data of the same mouse brain slice through multi-omics data fusion, generating spatial metabolic data corresponding to the coordinates of the spatial transcriptomic data, resulting in the second derived data.

[0248] Figure 19 This is a schematic diagram of the experimental results of multi-omics data integration according to an embodiment of this disclosure. Figure 19 This presents the results of a multi-omics data integration experiment on mouse brain tissue slices with sample number V11L12-038_B1. The spatial transcriptomic and spatial metabolomic data of the mouse brain tissue slices with sample number V11L12-038_B1 were integrated using the STAGE model to generate the first derived data. The spatial transcriptomic and spatial metabolomic data of the mouse brain tissue slices with sample number V11L12-038_B1 were then integrated using the SM2ST model to generate the second derived data.

[0249] like Figure 19 As shown, ion intensity distribution diagram 191 shows the ion intensity distribution with a mass-to-charge ratio of 522.35 in the original data. Ion intensity distribution diagram 192 shows the ion intensity distribution with a mass-to-charge ratio of 522.35 in the first derived data. Ion intensity distribution diagram 193 shows the ion intensity distribution with a mass-to-charge ratio of 522.35 in the second derived data.

[0250] like Figure 19As shown in Venn diagram 194, the overlap between the original set C11 and the first set C12 is illustrated. The original set C11 represents the set of highly variable metabolic ions in the original data, while the first set C12 represents the set of highly variable metabolic ions in the first derived data. The original set C11 includes regions B11 and B12, and the first set C12 includes regions B12 and B13. In other words, region B12 is the intersection of the original set C11 and the first set C12. Region B11 contains 33 metabolic ions, region B12 contains 917, and region B13 contains 1061. Regression diagram 195 shows the linear regression results of highly variable metabolic ions in the original data and the first derived data, demonstrating their linear correlation. .

[0251] like Figure 19 As shown in Venn diagram 196, the overlap between the original set C11 and the second set C13 is illustrated. The original set C11 represents the set of highly variable metabolic ions in the original data, while the second set C13 represents the set of highly variable metabolic ions in the second derived data. The original set C11 includes regions B21 and B22, and the second set C13 includes regions B22 and B23. In other words, region B22 is the intersection of the original set C11 and the second set C13. Region B21 contains 16 metabolic ions, region B22 contains 934, and region B23 contains 1093. Regression diagram 197 shows the linear regression results of highly variable metabolic ions in the original data and the second derived data, demonstrating their linear correlation. .

[0252] Figure 20 This is a schematic diagram of the experimental results of multi-omics data integration according to another embodiment of this disclosure. Figure 20 This presents the results of a multi-omics data integration experiment on mouse brain tissue slices with sample number V11L12-038_A1. The spatial transcriptomic and spatial metabolomic data of the mouse brain tissue slices with sample number V11L12-038_A1 were integrated using the STAGE model to generate the first derived data. The spatial transcriptomic and spatial metabolomic data of the mouse brain tissue slices with sample number V11L12-038_A1 were then integrated using the SM2ST model to generate the second derived data.

[0253] like Figure 20As shown, ion intensity distribution map 201 shows the ion intensity distribution with a mass-to-charge ratio of 367.32 in the original data. Ion intensity distribution map 202 shows the ion intensity distribution with a mass-to-charge ratio of 367.32 in the first derived data. Ion intensity distribution map 203 shows the ion intensity distribution with a mass-to-charge ratio of 367.32 in the second derived data.

[0254] like Figure 20 As shown in Venn diagram 204, the overlap between the original set C21 and the first set C22 is illustrated. The original set C21 represents the set of highly variable metabolic ions in the original data, while the first set C22 represents the set of highly variable metabolic ions in the first derived data. The original set C21 includes regions B31 and B32, and the first set C22 includes regions B32 and B33. In other words, region B32 is the intersection of the original set C21 and the first set C22. Region B31 contains 17 metabolic ions, region B32 contains 764, and region B33 contains 1201. Regression diagram 205 shows the linear regression results of highly variable metabolic ions in the original data and the first derived data, demonstrating their linear correlation. .

[0255] like Figure 20 As shown in Venn diagram 206, the overlap between the original set C21 and the second set C23 is illustrated. The original set C21 represents the set of highly variable metabolic ions in the original data, while the second set C23 represents the set of highly variable metabolic ions in the second derived data. The original set C21 includes regions B41 and B42, and the second set C23 includes regions B42 and B43. In other words, region B42 is the intersection of the original set C21 and the second set C23. Region B41 contains 4 metabolic ions, region B42 contains 777 metabolic ions, and region B43 contains 1222 metabolic ions. Regression diagram 207 shows the linear regression results of highly variable metabolic ions in the original data and the second derived data, demonstrating their linear correlation. .

[0256] Figure 21 This is a schematic diagram of the experimental results of multi-omics data integration according to another embodiment of this disclosure. Figure 21This presents the results of a multi-omics data integration experiment on mouse brain tissue slices with sample number V11L12-109_B1. The spatial transcriptomic and spatial metabolomic data of the mouse brain tissue slices with sample number V11L12-109_B1 were integrated using the STAGE model to generate the first derived data. The spatial transcriptomic and spatial metabolomic data of the mouse brain tissue slices with sample number V11L12-109_B1 were then integrated using the SM2ST model to generate the second derived data.

[0257] like Figure 21 As shown, ion intensity distribution diagram 211 shows the ion intensity distribution with a mass-to-charge ratio of 573.22 in the original data. Ion intensity distribution diagram 212 shows the ion intensity distribution with a mass-to-charge ratio of 573.22 in the first derived data. Ion intensity distribution diagram 213 shows the ion intensity distribution with a mass-to-charge ratio of 573.22 in the second derived data.

[0258] like Figure 21 As shown in Venn diagram 214, the overlap between the original set C31 and the first set C32 is illustrated. The original set C31 represents the set of highly variable metabolic ions in the original data, while the first set C32 represents the set of highly variable metabolic ions in the first derived data. The original set C31 includes regions B51 and B52, and the first set C32 includes regions B52 and B53. In other words, region B52 is the intersection of the original set C31 and the first set C32. Region B51 contains 2 metabolic ions, region B52 contains 33 metabolic ions, and region B53 contains 304 metabolic ions. Regression diagram 215 shows the linear regression results of highly variable metabolic ions in the original data and the first derived data, demonstrating their linear correlation. .

[0259] like Figure 21 As shown in Venn diagram 216, the overlap between the original set C31 and the second set C33 is illustrated. The original set C31 represents the set of highly variable metabolic ions in the original data, while the second set C33 represents the set of highly variable metabolic ions in the second derived data. The original set C31 includes regions B61 and B62, and the second set C33 includes regions B62 and B63. In other words, region B62 is the intersection of the original set C31 and the second set C33. Region B61 contains 0 metabolic ions, region B62 contains 35, and region B63 contains 317. Regression diagram 217 shows the linear regression results of highly variable metabolic ions in the original data and the second derived data, demonstrating their linear correlation. .

[0260] Figure 22 This is a schematic diagram of the experimental results of multi-omics data integration according to another embodiment of this disclosure. Figure 22 This presents the results of a multi-omics data integration experiment on mouse brain tissue slices with sample number V11L12-109_C1. The spatial transcriptomic and spatial metabolomic data of the mouse brain tissue slices with sample number V11L12-109_C1 were integrated using the STAGE model to generate the first derived data. The spatial transcriptomic and spatial metabolomic data of the mouse brain tissue slices with sample number V11L12-109_C1 were then integrated using the SM2ST model to generate the second derived data.

[0261] like Figure 22 As shown, ion intensity distribution diagram 221 shows the ion intensity distribution with a mass-to-charge ratio of 674.28 in the original data. Ion intensity distribution diagram 222 shows the ion intensity distribution with a mass-to-charge ratio of 674.28 in the first derived data. Ion intensity distribution diagram 223 shows the ion intensity distribution with a mass-to-charge ratio of 674.28 in the second derived data.

[0262] like Figure 22 As shown in Venn diagram 224, the overlap between the original set C41 and the first set C42 is illustrated. The original set C41 represents the set of highly variable metabolic ions in the original data, while the first set C42 represents the set of highly variable metabolic ions in the first derived data. The original set C41 includes regions B71 and B72, and the first set C42 includes regions B72 and B73. In other words, region B72 is the intersection of the original set C41 and the first set C42. Region B71 contains 1 metabolic ion, region B72 contains 42 metabolic ions, and region B73 contains 360 metabolic ions. Regression diagram 225 shows the linear regression results of highly variable metabolic ions in the original data and the first derived data, demonstrating their linear correlation. .

[0263] like Figure 22 As shown in Venn diagram 226, the overlap between the original set C41 and the second set C43 is illustrated. The original set C41 represents the set of highly variable metabolic ions in the original data, while the second set C43 represents the set of highly variable metabolic ions in the second derived data. The original set C41 includes regions B81 and B82, and the second set C43 includes regions B82 and B83. In other words, region B82 is the intersection of the original set C41 and the second set C43. Region B81 contains 0 metabolic ions, region B82 contains 43, and region B83 contains 385. Regression diagram 227 shows the linear regression results of highly variable metabolic ions in the original data and the second derived data, illustrating their linear correlation. .

[0264] Figure 23 This is a schematic diagram illustrating the experimental results of multi-omics data integration according to yet another embodiment of this disclosure. Figure 19 same, Figure 23 It is also the result of a multi-omics data integration experiment of mouse brain tissue slices with sample number V11L12-038_B1. Figure 19 Region B21 is a set of highly variable metabolic ions that exist in the original data but were not generated in the second derived data, including metabolic ions with mass-to-charge ratios of 880.05, 614.18 and 793.18.

[0265] It should be noted that since the metabolic ions in region B21 are externally expressed, such as the metabolic ions corresponding to mass-to-charge ratios of 880.05, 614.18 and 793.18, the SM2ST model failed to reduce these metabolic ions.

[0266] like Figure 23 As shown, ion intensity distribution diagram 231 shows the ion intensity distribution with a mass-to-charge ratio of 880.05 in the original data, and ion intensity distribution diagram 232 shows the ion intensity distribution with a mass-to-charge ratio of 880.05 in the second derived data.

[0267] like Figure 23 As shown, ion intensity distribution diagram 233 shows the ion intensity distribution with a mass-to-charge ratio of 614.18 in the original data, and ion intensity distribution diagram 234 shows the ion intensity distribution with a mass-to-charge ratio of 614.18 in the second derived data.

[0268] like Figure 23 As shown, ion intensity distribution diagram 235 shows the ion intensity distribution with a mass-to-charge ratio of 793.18 in the original data, and ion intensity distribution diagram 236 shows the ion intensity distribution with a mass-to-charge ratio of 793.18 in the second derived data.

[0269] Figure 24 This is a schematic diagram illustrating the experimental results of multi-omics data integration according to yet another embodiment of this disclosure. Figure 19 same, Figure 24 It is also the result of a multi-omics data integration experiment of mouse brain tissue slices with sample number V11L12-038_B1. Figure 19 Region B23 in the data is a set of highly variable metabolic ions that are not present in the original data but are generated in the second derived data, including metabolic ions with mass-to-charge ratios of 735.87, 823.54 and 557.90.

[0270] It should be noted that the metabolic ions in region B23 were not considered highly variable metabolic ions in the original data. However, in the second derived data, the SM2ST model was able to restore the signals of these metabolic ions to the same location with lower noise and enhanced tissue specificity of the ion signals. Therefore, after the data was restored by the SM2ST model, they were considered highly variable metabolic ions.

[0271] like Figure 24 As shown, ion intensity distribution diagram 241 shows the ion intensity distribution with a mass-to-charge ratio of 735.87 in the original data, and ion intensity distribution diagram 242 shows the ion intensity distribution with a mass-to-charge ratio of 735.87 in the second derived data.

[0272] like Figure 24 As shown, ion intensity distribution diagram 243 shows the ion intensity distribution with a mass-to-charge ratio of 823.54 in the original data, and ion intensity distribution diagram 244 shows the ion intensity distribution with a mass-to-charge ratio of 823.54 in the second derived data.

[0273] like Figure 24 As shown, ion intensity distribution diagram 245 shows the ion intensity distribution with a mass-to-charge ratio of 557.90 in the original data, and ion intensity distribution diagram 246 shows the ion intensity distribution with a mass-to-charge ratio of 557.90 in the second derived data.

[0274] It should be noted here that, Figures 19 to 24 In the multi-omics data integration experiment, the original data, the first derived data, and the second derived data were all based on... and ,as well as and Conditional screening of highly variable metabolic ions. Based on the results of the multi-omics data integration experiment, the STAGE model cannot effectively reconstruct metabolic signals highly expressed in the corpus callosum. According to the linear regression results, the linear correlation between the first and second derived data and the original data both exceeded 98%, but the boxes near 0 (e.g.) Figure 19 Box Q6 in the middle Figure 20 Box Q7 in the middle Figure 21 The box Q8 and Figure 22 In box Q9), the data generated by the STAGE model cannot reconstruct low-intensity, spatially highly variable metabolic signals. According to the Venn diagram, the SM2ST model retains a higher number of ion signals for spatially highly variable ions than the STAGE model. For example... Figure 21 As shown, the STAGE model not only fails to preserve low-signal spatially highly variable metabolic signals, but also fails to clearly reconstruct the expression locations of metabolic ion signals, resulting in lower linear correlation.

[0275] It should also be noted here that, as Figure 22 As shown, in the mouse brain tissue slice with sample number V11L12-109_C1, the right side was treated with unilateral 6-hydroxydopamine (6-OHDA)-lesioned, resulting in the gradual loss of dopaminergic neurons. The right caudoputamen (CP) showed low metabolic signal intensity, while the left CP showed high metabolic signal intensity. Only the SM2ST model could accurately reconstruct this. The autoencoder provides data noise reduction capabilities, mitigating noise points caused by experimental factors (e.g., drift during long-term instrument acquisition, uneven matrix application, and characteristics of the quality analyzer). Thanks to the adversarial structure of the SM2ST model, the network's generalization ability is enhanced by generating richer coordinate information, and highly variable metabolic signals are better preserved.

[0276] In some embodiments, spatial transcriptomic and spatial metabolomic data from mouse brain tissue slices are used for clustering to obtain clustering experimental results.

[0277] In some embodiments, unilateral dopaminergic neurons of mice to be treated are processed with 6-hydroxydopamine (6-OHDA)-lesioned to obtain a first group of mice. Brain tissue slices are obtained from the first group of mice to obtain first group mouse brain tissue slices. Brain tissue slices are obtained from normal mice to obtain a second group of mouse brain tissue slices. For each mouse brain tissue slice in the first and second groups, spatial metabolomics data and spatial transcriptomics data are obtained based on the SMA protocol; integrated data of spatial metabolomics data and spatial transcriptomics data are obtained using the spatial metabolic data acquisition method described in the above embodiments; clustering algorithms are used to cluster the spatial metabolomics data, spatial transcriptomics data, and integrated data respectively to obtain the clustering results of spatial metabolomics data, spatial transcriptomics data, and integrated data.

[0278] For example, Spatialglue can be used for clustering. Spatialglue is a tool that supports spatial multi-omics clustering, aligning data from different patterns to the same latent space. Using combined spatial metabolomics and spatial transcriptomics data from mouse brain tissue slices as the research object, multimodal clustering analysis was performed, including: screening for highly variable genes in the spatial omics data of the mouse brain, and performing Principal Component Analysis (PCA) on the screened data, setting the dimension to 50. Parameter settings included: datatype set to 10X, learning_rate set to 0.0001, weight_decay set to 0.00001, and epochs set to 200.

[0279] Figure 25 and Figure 26 This is a schematic diagram of the clustering experiment results of one embodiment of the present disclosure. Figure 25 and Figure 26 The results are from a clustering experiment of a mouse brain tissue slice with sample number V11L12-038_B1, which was untreated.

[0280] like Figure 25 As shown, the spatial transcriptome data 251 is clustered using a clustering algorithm, resulting in the first clustering result 252. The spatial metabolome data 253 is clustered using the same algorithm, resulting in the second clustering result 254. The integrated data 255 is then clustered using the same algorithm, resulting in the third clustering result 256. The third clustering result 256 includes multiple categories, where the modality weights for each category are as follows: Figure 26 As shown.

[0281] like Figure 26 As shown, the horizontal axis represents the multiple categories in the third clustering result 256, and the vertical axis represents the modality weights. For example, curve 261 is the spatial transcriptome data weight curve for category 4 in the third clustering result 256, and curve 262 is the spatial metabolome data weight curve for category 4 in the third clustering result 256.

[0282] Figure 27 and Figure 28 This is a schematic diagram of the clustering experiment results of another embodiment of this disclosure. Figure 27 and Figure 28 The results are from a clustering experiment of a mouse brain tissue slice with sample number V11L12-038_A1, which was untreated.

[0283] like Figure 27As shown, the spatial transcriptome data 271 is clustered using a clustering algorithm, resulting in the first clustering result 272. The spatial metabolome data 273 is clustered using the same algorithm, resulting in the second clustering result 274. The integrated data 275 is then clustered using the same algorithm, resulting in the third clustering result 276. The third clustering result 276 includes multiple categories, where the modality weights for each category are as follows: Figure 28 As shown.

[0284] like Figure 28 As shown, the horizontal axis represents the multiple categories in the third clustering result 276, and the vertical axis represents the modality weights. For example, curve 281 is the spatial transcriptome data weight curve for category 9 in the third clustering result 276, and curve 282 is the spatial metabolome data weight curve for category 9 in the third clustering result 276.

[0285] Figure 29 and Figure 30 This is a schematic diagram of the clustering experiment results of another embodiment of this disclosure. Figure 29 and Figure 30 The results are from a clustering experiment of a mouse brain tissue slice with sample number V11L12-109_B1, which was subjected to unilateral 6-hydroxydopamine (6-OHDA)-lesioned damage to dopaminergic neurons.

[0286] like Figure 29 As shown, the spatial transcriptome data 291 is clustered using a clustering algorithm, resulting in the first clustering result 292. The spatial metabolome data 293 is clustered using the same algorithm, resulting in the second clustering result 294. The integrated data 295 is then clustered using the same algorithm, resulting in the third clustering result 296. The third clustering result 296 includes multiple categories, where the modality weights for each category are as follows: Figure 30 As shown.

[0287] like Figure 30 As shown, the horizontal axis represents the multiple categories in the third clustering result 296, and the vertical axis represents the modality weights. For example, curve 301 is the spatial transcriptome data weight curve for category 12 in the third clustering result 296, and curve 302 is the spatial metabolome data weight curve for category 12 in the third clustering result 296.

[0288] Figure 31 and Figure 32 This is a schematic diagram of the clustering experiment results of another embodiment of this disclosure. Figure 31 and Figure 32 The results are from a clustering experiment of a mouse brain tissue slice with sample number V11L12-109_C1, which was subjected to unilateral 6-hydroxydopamine (6-OHDA)-lesioned damage to dopaminergic neurons.

[0289] like Figure 31 As shown, the spatial transcriptome data 3101 is clustered using a clustering algorithm, resulting in a first clustering result 3102. The spatial metabolome data 3103 is clustered using the same algorithm, resulting in a second clustering result 3104. The integrated data 3105 is then clustered using the same algorithm, resulting in a third clustering result 3106. The third clustering result 3106 includes multiple categories, where the modality weights for each category are as follows: Figure 32 As shown.

[0290] like Figure 32 As shown, the horizontal axis represents the multiple categories in the third clustering result 3106, and the vertical axis represents the modality weights. For example, curve 3201 is the spatial transcriptome data weight curve for category 6 in the third clustering result 3106, and curve 3202 is the spatial metabolome data weight curve for category 6 in the third clustering result 3106.

[0291] It should be noted here that, Figures 25 to 32 In clustering experiments, the SpatialGlue spatial multi-omics clustering algorithm was used for clustering. It was found that regardless of whether the mice were treated, in the results of clustering using only transcriptome data, the two CPs were classified into the same category (e.g., Figure 25 The first clustering result in the dataset is 252. Figure 27 The first clustering result in the dataset is 272. Figure 29 The first clustering result 292 and Figure 31 The first clustering result (3102) shows that in the results of clustering using only metabolic information, the CPs on both sides are divided into different categories (e.g. Figure 25 The second clustering result 254 in the data. Figure 27 The second clustering result 274 in the data. Figure 29 The second clustering result 294 and Figure 31 As shown in the second clustering result 3104), single modality information can easily lead to incorrect tissue classification. In the clustering results obtained by introducing multimodal information, the CPs on both sides of the untreated mouse brain were classified into the same category (e.g., Figure 25 The third clustering result 256 and Figure 27 As shown in the third clustering result (276), the CPs on both sides of the brain of the treated mice were divided into different categories (e.g., Figure 29 The third clustering result 296 and Figure 31 The third clustering result (3106) is shown in the figure, which is consistent with the fact that only the left CP expressed dopamine neural information in the experiment.

[0292] It should also be noted that further examination of the modal weights learned by SpatialGlue revealed that, due to the greater impact of the gradual loss of dopaminergic neurons on metabolic signaling, the transcriptomic modality in the CP region of the brain of mice treated with unilateral 6-hydroxydopamine (6-OHDA)-lesioned made a more significant contribution than the metabolomic modality. For example, Figure 29 and Figure 30 In the middle, the treated side is like Figure 29 As shown in box Q11, this corresponds to category 10 in the third clustering result 296. In the modality weights of category 10 in the third clustering result 296, the weight of the transcriptome modality is greater than that of the metabolome modality (e.g., ...). Figure 30 (As shown in box Q14). For example, Figure 31 and Figure 32 In the middle, the treated side is like Figure 31 As shown in box Q15, this corresponds to category 3 in the third clustering result 3106. In the modality weights of category 3 in the third clustering result 3106, the weight of the transcriptome modality is greater than that of the metabolome modality (e.g., ...). Figure 32 (as shown in box Q18).

[0293] It should also be noted that, conversely, the untreated metabolomics modality made a more significant contribution than the transcriptomics modality. For example, Figure 29 and Figure 30 In the middle, the untreated side is like... Figure 29 As shown in box Q12, this corresponds to category 3 in the third clustering result 296. In the modality weights of category 3 in the third clustering result 296, the weight of the metabolomics modality is greater than that of the transcriptomics modality (e.g., ...). Figure 30 (As shown in box Q13). For example, Figure 31 and Figure 32 In the middle, the untreated side is like... Figure 31 As shown in box Q16, this corresponds to categories 2 and 15 in the third clustering result 3106. In the modality weights of categories 2 and 15 in the third clustering result 3106, the weight of the metabolomics modality is greater than that of the transcriptomics modality (e.g., ...). Figure 32 (as shown in boxes Q17 and Q19).

[0294] Figure 33 and Figure 34 This is a schematic diagram illustrating the clustering experiment results of yet another embodiment of this disclosure. Figure 31 and Figure 32 same, Figure 33 and Figure 34 This is also the result of a clustering experiment on mouse brain tissue slices with sample number V11L12-109_C1. The treated side, as shown... Figure 31 As shown in box Q15, it corresponds to category 3 in the third clustering result 3106. The unprocessed side is shown below. Figure 31 As shown in box Q16, this corresponds to category 15 in the third clustering result 3106. A difference analysis was performed on categories 3 and 15 in the third clustering result 3106, and the results are as follows: Figure 33 As shown, this includes differences in the expression levels of the Cplx2, Pcp4, Egr, Dgkb, and Tac1 genes. According to... Figure 33 The differential analysis results yielded the expression level distribution of each gene, such as... Figure 34 As shown.

[0295] like Figure 33 As shown, the horizontal axis represents gene type, and the vertical axis represents expression level. For example, curve 3301 is the expression level curve of the Pcp4 gene in category 3, and curve 3302 is the expression level curve of the Pcp4 gene in category 15.

[0296] like Figure 34 As shown, expression distribution diagram 3401 shows the expression distribution of the Cplx2 gene, expression distribution diagram 3402 shows the expression distribution of the Pcp4 gene, expression distribution diagram 3403 shows the expression distribution of the Egr gene, expression distribution diagram 3404 shows the expression distribution of the Dgkb gene, and expression distribution diagram 3405 shows the expression distribution of the Tac1 gene. Figure 34 Boxes Q20, Q21, Q22, Q23, and Q25 in the diagram represent the processed side.

[0297] For example, the Wilcoxon rank-sum test in scanpy.tl.Rank_genes_groups can be used as a statistical method to perform differential analysis, thereby comparing gene expression differences between two clusters.

[0298] It should be noted here that, according to Figure 33 and Figure 34 It can be seen that the genes Cplx2, Pcp4, Egr1, Dgkb and Tac1 on the undamaged side are all relatively highly expressed in CP, which indicates that these genes are positively correlated with dopamine expression.

[0299] The clustering results above demonstrate that learning data from two modalities simultaneously can more accurately characterize the state of the mouse brain.

[0300] Figure 35 This is a schematic diagram illustrating the result of a high-density spatial metabolic data generation method according to an embodiment of this disclosure. Figure 35 In China, with Spatial metabolic data of mouse brain tissue slices were acquired using spatial resolution to obtain high-resolution data. Spatial metabolic data from the same mouse brain tissue slice was acquired using a spatial resolution method, resulting in low-resolution data. The low-resolution data was then processed using the high-density spatial metabolic data generation method described in any of the above embodiments to obtain super-resolution data.

[0301] like Figure 35 As shown, ion intensity distribution map 3501 shows the ion intensity distribution with a mass-to-charge ratio of 716.45 in high-resolution data, ion intensity distribution map 3502 shows the ion intensity distribution with a mass-to-charge ratio of 716.45 in low-resolution data, and ion intensity distribution map 3503 shows the ion intensity distribution with a mass-to-charge ratio of 716.45 in super-resolution data.

[0302] like Figure 35 As shown, ion intensity distribution map 3504 shows the ion intensity distribution with a mass-to-charge ratio of 837.67 in high-resolution data, ion intensity distribution map 3505 shows the ion intensity distribution with a mass-to-charge ratio of 837.67 in low-resolution data, and ion intensity distribution map 3506 shows the ion intensity distribution with a mass-to-charge ratio of 837.67 in super-resolution data.

[0303] like Figure 35 As shown, ion intensity distribution map 3507 shows the ion intensity distribution with a mass-to-charge ratio of 874.5 in high-resolution data, ion intensity distribution map 3508 shows the ion intensity distribution with a mass-to-charge ratio of 874.5 in low-resolution data, and ion intensity distribution map 3509 shows the ion intensity distribution with a mass-to-charge ratio of 874.5 in super-resolution data.

[0304] like Figure 35 As shown, clustering of high-resolution data yields high-resolution clustering result 3510. Clustering of low-resolution data yields low-resolution clustering result 3511. Clustering of super-resolution data yields super-resolution clustering result 3512.

[0305] It's important to note that STMGraph is a versatile spatial omics method with reconstruction and clustering capabilities. In low-resolution imaging, due to the strong laser and large speckle size, each speckle covers a wide area, resulting in insufficient detail in the metabolic information represented by each pixel. Conversely, in high-resolution imaging, the weaker laser and smaller speckle size lead to the loss of some ion information. To address this issue, a pixel can be subdivided into four sites to construct a spatial neighborhood map, and then STMGraph can be used to reconstruct each pixel.

[0306] It should also be noted here that, according to Figure 35It is evident that some high mass-to-charge ratio ion signals are weaker in high-resolution imaging (as shown in ion intensity distribution maps 3501, 3504, and 3507), but have higher signal intensity in low-resolution imaging (as shown in ion intensity distribution maps 3502, 3505, and 3508). Mass-to-charge ratios of 716.45 and 874.50 are preserved after super-resolution imaging (as shown in ion intensity distribution maps 3501, 3503, 3507, and 3509), and the signal of a mass-to-charge ratio of 837.67 in the hippocampal region is enhanced (as shown in ion intensity distribution maps 3504 and 3506). This is attributed to the less excitation damage in low-resolution imaging; utilizing low-resolution information for high-density spatial imaging better preserves ion information.

[0307] It should also be noted that low-resolution metabolic information cannot adequately describe the hippocampus region of the mouse brain (as shown in high-resolution clustering result 3510 and low-resolution clustering result 3511). High-density imaging, however, can more clearly preserve the characteristics of the hippocampus region (as shown in super-resolution clustering result 3512). This demonstrates that STMGraph has a significant advantage in improving the resolution and accuracy of spatial metabolomics data.

[0308] Figure 36 This is a schematic diagram of the structure of a spatial metabolic data acquisition device according to an embodiment of the present disclosure.

[0309] like Figure 36 As shown, the space metabolic data acquisition device 360 ​​can be represented in the form of a general-purpose computing device. The space metabolic data acquisition device 360 ​​includes a memory 361, a processor 362, a bus 363, an input / output interface 364, a network interface 365, and a storage interface 366. Figure 7 The difference is that processor 362 is configured to execute instructions stored in memory 361, such as... Figure 8 The spatial metabolic data acquisition method involved in any of the embodiments.

[0310] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations thereof, can be implemented by computer-readable program instructions.

[0311] These computer-readable program instructions are provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable device to produce a machine, such that execution of the instructions by the processor produces means for implementing the functions specified in one or more boxes of the flowchart and / or block diagram.

[0312] These computer-readable program instructions may also be stored in a computer-readable storage medium. These instructions cause a computer to work in a particular manner to produce an article of manufacture, including instructions that implement the functions specified in one or more boxes in a flowchart and / or block diagram.

[0313] This disclosure may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.

[0314] This disclosure also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement... Figure 8 The method involved in any of the embodiments.

[0315] This disclosure also provides a computer program product, including computer instructions, wherein the computer instructions, when executed by a processor, implement as follows: Figure 8 The method involved in any of the embodiments.

[0316] The beneficial effects obtained by implementing the above embodiments of this disclosure are as follows:

[0317] (1) The SM2ST model effectively reduces the noise of the original data by using a generative adversarial network model, while retaining highly variable metabolic ions and enhancing the spatial specificity of insignificant signals.

[0318] (2) Combining the STMGraph link prediction model, the SM2ST model can construct high-density spatial metabolic data, improve the clarity of ion signals with mass-to-charge ratio, and significantly improve the accuracy of spatial metabolomics clustering.

[0319] (3) The SM2ST model provides strong support for users by constructing anndata data, including a large user community and rich documentation resources, thereby providing new biomarkers for disease diagnosis from a multi-omics perspective and providing deeper insights for the study of cell behavior and tissue function.

[0320] In some embodiments, the functional units described above may be implemented as general-purpose processors, programmable logic controllers (PLCs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any suitable combination thereof for performing the functions described herein.

[0321] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0322] The description in this disclosure is provided for illustrative and descriptive purposes only and is not intended to be exhaustive or to limit the disclosure to its forms. Many modifications and variations will be apparent to those skilled in the art. The embodiments were chosen and described in order to better illustrate the principles and practical application of this disclosure and to enable those skilled in the art to understand this disclosure and to design various embodiments with various modifications suitable for a particular purpose.

Claims

1. A model training method, comprising: Mass spectrometry imaging was performed on biological tissue slices to obtain mass spectrometry image samples; The biological tissue section samples were stained with hematoxylin and eosin (HE) to obtain stained image samples; The mass spectrometry image sample is registered with the stained image sample to obtain the target image sample; Acquire spatial metabolic data for each point in a specified point set within the target image sample; The spatial metabolic data of each point is processed using an encoder in a generative adversarial network model to obtain the predicted coordinates of each point; The predicted coordinates of each point are processed using the decoder in the generative adversarial network model to obtain the spatial metabolic reconstruction data of each point; The probability that the coordinates of each point are true coordinates is identified by the discriminator in the generative adversarial network model, which is used as the confidence level of the coordinates of each point. The probability that the predicted coordinates of each point are the true coordinates is identified by the discriminator, and this probability is used as the confidence level of the predicted coordinates of each point. The loss value is determined based on the spatial metabolic data of each point, the spatial metabolic reconstruction data of each point, the coordinates of each point, the predicted coordinates of each point, the confidence level of the coordinates of each point, and the confidence level of the predicted coordinates of each point. The encoder, the decoder, and the discriminator are trained using the loss value.

2. The model training method according to claim 1, wherein determining the loss value includes: Based on the spatial metabolic data and the spatial metabolic reconstruction data of each point, determine the first sub-loss value; The second sub-loss value is determined based on the coordinates of each point, the predicted coordinates of each point, the confidence level of the coordinates of each point, and the confidence level of the predicted coordinates of each point. The loss value is obtained based on the first sub-loss value and the second sub-loss value.

3. The model training method according to claim 2, wherein determining the first sub-loss value includes: Calculate the weighted mean square error of the spatial metabolic data and the spatial metabolic reconstruction data of each point to obtain the first deviation value; The mean absolute error between the spatial metabolic data and the spatial metabolic reconstruction data at each point is calculated to obtain the second deviation value. The first sub-loss value is obtained by calculating the weighted sum of the first deviation value and the second deviation value.

4. The model training method according to claim 3, wherein, The spatial metabolic data of the i-th point is a data vector comprising M elements, wherein the j-th element of the data vector represents the metabolic intensity of the i-th point at the j-th mass-to-charge ratio. N is the total number of points in the specified point set. M represents the total number of mass-to-charge ratios.

5. The model training method according to claim 4 further includes: Determining the weight of the first deviation value, wherein determining the weight of the first deviation value includes: Construct a metabolic intensity matrix, wherein the metabolic intensity matrix includes spatial metabolic data of N points in the specified point set; Calculate the sum of all metabolic intensities for the M mass-to-charge ratio values ​​in the metabolic intensity matrix to obtain the matrix sum value; Calculate the sum of all metabolic intensities for the j-th mass-to-charge ratio value in the metabolic intensity matrix to obtain the j-th sum value; Calculate the ratio of the j-th sum to the sum of the matrix to obtain the j-th ratio; Calculate the difference between the predetermined value and the j-th ratio to obtain the j-th weight value in the weight of the first deviation value.

6. The model training method according to claim 2, wherein determining the second sub-loss value includes: Calculate the average absolute error between the coordinates of each point and the predicted coordinates of each point to obtain the coordinate error value; Calculate the average confidence level of the coordinates of each point to obtain the first confidence level mean. Calculate the average confidence level of the predicted coordinates of each point to obtain the second confidence level mean. The intermediate loss value is obtained by calculating the weighted sum of the coordinate error value and the mean of the first confidence level; The difference between the intermediate loss value and the second confidence mean is calculated to obtain the second sub-loss value.

7. The model training method according to claim 2, wherein obtaining the loss value based on the first sub-loss value and the second sub-loss value includes: The weighted sum of the first sub-loss value and the second sub-loss value is calculated to obtain the loss value.

8. The model training method according to claim 1, further comprising: Select multiple points to add within the neighborhood of each point in the specified point set; The spatial metabolic data of the points in the specified set of points closest to the added point are used as the spatial metabolic data of the added point. The spatial metabolic data of the kth addition point is updated using the spatial metabolic data of its eight neighboring points. K is the total number of points to be added to the specified point set.

9. The model training method according to claim 8, wherein selecting multiple additional points within the neighborhood of each point in the specified point set comprises: A polar coordinate system is established for each point, wherein each point is the pole of the polar coordinate system, and the ray extending from the pole along a predetermined direction is the polar axis of the polar coordinate system; The corresponding points of multiple candidate coordinates in the polar coordinate system are used as multiple addition points.

10. The model training method according to claim 9, wherein, The plurality of candidate coordinates includes a first candidate coordinate. Second candidate coordinates and the third candidate coordinates , where R is the predetermined distance.

11. The model training method according to any one of claims 1-10, wherein registering the mass spectrometry image sample with the stained image sample to obtain the target image sample comprises: Spatial metabolic data with highly spatially variable mass-to-charge ratio values ​​are extracted from the mass spectrometry image samples to generate image samples to be processed. Using an affine transformation matrix, each pixel in the image sample to be processed is mapped to the stained image coordinate system to obtain the target image sample.

12. The model training method according to claim 11, further comprising: The original transformation matrix is ​​trained to obtain the affine transformation matrix, wherein the trained original transformation matrix includes: The original transformation matrix is ​​generated based on the horizontal scaling factor, vertical scaling factor, horizontal shearing factor, vertical shearing factor, horizontal translation amount, and vertical translation amount. Using the original transformation matrix, the first coordinates of each pixel in the image sample to be processed are transformed into the stained image sample to obtain the second coordinates of each pixel; Extract the coordinates of feature points in the stained image sample that have a one-to-one correspondence with each pixel; Calculate the mean square error of the coordinates of the second coordinates of each pixel and the coordinates of the feature points that correspond one-to-one with each pixel; Based on the mean square error of the coordinates, the horizontal scaling factor, the vertical scaling factor, the horizontal shearing factor, the vertical shearing factor, the horizontal translation amount, and the vertical translation amount are trained to obtain the affine transformation matrix.

13. A model training device, comprising: Memory; A processor, coupled to a memory, is configured to implement the model training method as described in any one of claims 1-12 based on memory-stored instruction execution.

14. A method for acquiring spatial metabolic data, comprising: Mass spectrometry imaging was performed on the biological tissue sections to be processed to obtain mass spectrometry images; The biological tissue sections to be processed were stained with hematoxylin and eosin (HE) to obtain stained images; The mass spectrometry image is registered with the staining image to obtain the target image; Obtain the coordinates of each of the multiple measurement points in the target image; The coordinates of each measurement point are processed using a decoder to obtain spatial metabolic data for each measurement point, wherein the decoder is trained using the model training method described in any one of claims 1-12.

15. The spatial metabolic data acquisition method according to claim 14, wherein registering the mass spectrometry image with the staining image to obtain the target image comprises: Spatial metabolic data with highly spatially variable mass-to-charge ratio values ​​are extracted from the mass spectrometry image to generate an image to be processed. Using an affine transformation matrix, each pixel in the image to be processed is mapped to the coloring image coordinate system to obtain the target image.

16. A spatial metabolic data acquisition device, comprising: Memory; A processor, coupled to a memory, is configured to execute instructions stored in the memory to implement the spatial metabolic data acquisition method as described in any one of claims 14-15.

17. A computer-readable storage medium, wherein, A computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method as described in any one of claims 1-12 and 14-15.

18. A computer program product comprising computer instructions, wherein the computer instructions, when executed by a processor, implement the method as described in any one of claims 1-12, 14-15.

Images