A method, system, device and medium for dividing tissue regions based on spatial metabolomics data

Abstract

The application discloses a tissue region division method and system based on spatial metabolomics data, a device and a medium, and relates to the field of spatial metabolomics. The method comprises the following steps: converting spatial metabolomics data of a tissue slice into an expression matrix and a first adjacency matrix; cutting an H&E staining slice image into a plurality of image blocks; constructing and training a SpaMIC model; the SpaMIC model comprises a region division module and a feature extraction module; and based on the expression matrix, the first adjacency matrix and the image blocks, the tissue slice is subjected to region division by using the trained SpaMIC model. The application can realize efficient and accurate division of the spatial structure region of the tissue.

Description

Technical Field

[0001] This application relates to the field of spatial metabolomics, and in particular to a method, system, device and medium for tissue region segmentation based on spatial metabolomics data. Background Technology

[0002] Spatial metabolomics, as an emerging scientific research technology, has demonstrated its enormous potential in elucidating the spatial distribution of metabolites within cells and tissues. This technology can provide information on the distribution of metabolites in microscopic space, often exhibiting significant region specificity, which is crucial for revealing the fine-grained regulatory mechanisms of cellular metabolic activities. However, current analytical tools for spatial metabolomics data are still immature, particularly in terms of region segmentation and integrated analysis. On the one hand, the data format of spatial metabolomics data itself is limited, requiring different data preprocessing and feature extraction methods. On the other hand, due to the relatively limited ability of metabolomics data to reveal spatial heterogeneity, it is necessary to combine it with other omics data, such as transcriptomics data and optical image information, for joint analysis to obtain more accurate region segmentation and more comprehensive biological information. Summary of the Invention

[0003] The purpose of this application is to provide a method, system, device and medium for tissue region segmentation based on spatial metabolomics data, which can achieve efficient and accurate segmentation of the spatial structural regions of tissues.

[0004] To achieve the above objectives, this application provides the following solution:

[0005] Firstly, this application provides a method for tissue region delineation based on spatial metabolomics data, including:

[0006] A method for tissue region delineation based on spatial metabolomics data, characterized by comprising:

[0007] Spatial metabolomics data of tissue sections were converted into expression matrices and first adjacency matrices;

[0008] The H&E stained section image was cropped into multiple image blocks;

[0009] Construct and train a SpaMIC model; the SpaMIC model includes a region segmentation module and a feature extraction module;

[0010] Based on the expression matrix, the first adjacency matrix, and the image patch, the trained SpaMIC model is used to divide the tissue slice into regions.

[0011] Secondly, this application provides a tissue region partitioning system based on spatial metabolomics data, including:

[0012] The transformation module is used to convert spatial metabolomics data of tissue slices into expression matrices and first adjacency matrices.

[0013] The image patch cropping module is used to crop H&E stained slide images into multiple image patches;

[0014] The model building and training module is used to build and train the SpaMIC model; the SpaMIC model includes a region segmentation module and a feature extraction module.

[0015] The region segmentation module is used to segment the tissue slice into regions based on the expression matrix, the first adjacency matrix, and the image patch using a trained SpaMIC model.

[0016] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for tissue region division based on spatial metabolomics data.

[0017] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for tissue region division based on spatial metabolomics data.

[0018] According to the specific embodiments provided in this application, this application has the following technical effects:

[0019] This application provides a method, system, device, and medium for tissue region segmentation based on spatial metabolomics data. Spatial metabolomics data of tissue slices usually exhibit obvious region specificity. The SpaMIC model, by integrating deep learning, can efficiently process and analyze spatial metabolomics data, greatly improving the efficiency of data analysis and achieving efficient and accurate segmentation of spatial structural regions of tissues. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating a tissue region partitioning method based on spatial metabolomics data provided in an embodiment of this application;

[0022] Figure 2The diagram illustrates the results of processing the pig embryo dataset using the SpaMIC model in Case 1. (a) is the original H&E stained histological image, (b) is the spatial domain resolved by the SpaMIC model, (c) is the spatial domain resolved by the SpaMIC model using the default Louvain clustering method, (d) is the spatial domain resolved by the SmartGate method, and (e) is the spatial domain resolved by the GraphST method.

[0023] Figure 3 This is a schematic diagram showing the results of processing mouse kidney slices using the SpaMIC model in Case 2; where (a) is the original H&E stained histological image, (b) is the domain analysis result of spatial metabolomics data, (c) is the domain analysis result of H&E omics image data, and (d) is the result after mapping modification.

[0024] Figure 4 This is a schematic diagram of the results of processing mouse brain slices using the SpaMIC model in Case 3; where (a) is the spatial domain analyzed using only spatial metabolomics data, (b) is the spatial domain analyzed after integrating spatial transcriptomics features, (c) is the spatial domain analyzed after integrating H&E image anomalous omics features, and (d) is the spatial distribution map of dopamine analysis.

[0025] Figure 5 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] Based on existing technological limitations, this application utilizes optical microscopic images of tissue sections to obtain rich spatial texture details, which can be used as reference images. Combined with metabolite information from spatial metabolomics, this enables refined division of tissue spatial regions. Furthermore, within the tissue itself, integrating transcriptomics sequencing technology to analyze the significant heterogeneity in cell types, functions, and gene expression across different regions facilitates more accurate regional division and a more comprehensive interpretation of biological characteristics.

[0028] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] In one exemplary embodiment, such as Figure 1 As shown, a method for tissue region segmentation based on spatial metabolomics data is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is described using a server as an example, and includes the following steps S1 to S4. Wherein:

[0030] S1: Convert the spatial metabolomics data of tissue sections into expression matrices and first adjacency matrices.

[0031] Spatial metabolomics data were first transformed into an expression matrix X and a first adjacency matrix A. In the expression matrix X, rows represent pixels, and columns represent selected m / z values. The expression matrix was normalized and logarithmically transformed using the scanpy package. After normalization, the sum of expression values ​​within a single pixel is 1e. 4 The K-nearest neighbor algorithm is used to transform spatial location coordinates into an undirected adjacency network, where node i and node j are each other's k nearest neighbors if and only if k is 5 by default. In addition, a method for calculating the adjacency matrix A based on a predefined radius r of Euclidean distance is provided, allowing for flexible selection in different datasets.

[0032] S2: Crop the H&E stained slice image into multiple image blocks.

[0033] A fixed spacing is used to determine the center point, and an image block of pixels within the range of W×H of the center point is cropped. H and W are the height and width of the image block, respectively. In this embodiment, H and W are set to 224 (H and W are set to 112 in a specific dataset), and the fixed spacing is 200.

[0034] Alternatively, based on the spatial coordinates of the spatial omics data, an image patch of W×H pixels is cropped around each coordinate point of the corresponding H&E stained slide image, where H and W are the height and width of the image patch, respectively. In this embodiment, H and W are set to 256 (H and W are set to 112 in a specific dataset).

[0035] S3: Construct and train the SpaMIC model; the SpaMIC model includes a region segmentation module and a feature extraction module.

[0036] (1) Region Division Module

[0037] The region partitioning module is the initial processing step of the SpaMIC model. In order to effectively reduce the dimensionality and noise interference of the original data, this application introduces an iterative clustering method to extract metabolites that are representative of the region.

[0038] The region segmentation module employs a graph attention network autoencoder. As a deep learning network, the graph attention network autoencoder learns node relationships and reconstructs node features from a graph structure. It mainly consists of an encoder and a decoder. The encoder adopts the basic structure of the GAT_module module of the MSPTI model, and the decoder structure is symmetrical to the encoder structure. This application refers to such a complete graph attention network autoencoder as the SpaMIC_AE model. Its training process minimizes the reconstruction loss of the normalized expression as follows:

[0039]

[0040] Where, x i x′ represents the original matrix. i This represents the expression value of the matrix output by the SpaMIC_AE model.

[0041] The region partitioning module comprises two iterative training processes. In the first iteration, the SpaMIC_AE model (i.e., the region partitioning module) is trained using the sample expression matrix and the first sample adjacency matrix as input. These matrices are derived from spatial metabolomics data of sample tissue slices. Its input and output layers are equal to the number of features in the sample expression matrix, with 512-30 hidden layers. Training uses the Adam optimizer with a specific initial learning rate and training epoch. The low-dimensional vector output by the encoder is subjected to Louvain clustering using the scanpy package by default, and differential analysis is used to obtain marker molecules from different cell populations. For each cell population, the top 1000 feature molecules are extracted as columns to form a new sample expression matrix. The number of cells in the updated sample expression matrix is ​​the same as the original expression matrix. Furthermore, molecules of interest with known m / z mass-to-charge ratios can be added to the updated sample expression matrix to avoid missing regionally representative metabolite molecules during automated feature selection. In the second iteration of training, the updated sample expression matrix and the first sample adjacency matrix are used as input to train the SpaMIC_AE model. Its input and output layers are equal to the number of features in the updated sample expression matrix Y, and the remaining training parameters are the same as above. It should be noted that the processing in this module is not limited to spatial metabolomics data; similar effects are achieved with expression matrices from other spatial omics data, such as spatial transcriptomics data.

[0042] (2) Feature extraction module

[0043] The feature extraction module combines image information to refine and optimize the region segmentation results. The feature extraction module employs a ResNet-50 network structure. In this embodiment, a ResNet-50 network structure pre-trained on ImageNet21k is used to extract 2048-dimensional feature vectors from image patches. The resulting image feature representation matrix and second adjacency matrix serve as inputs to the region segmentation module.

[0044] S4: Based on the expression matrix, the first adjacency matrix, and the image patch, the trained SpaMIC model is used to divide the tissue slice into regions. Specifically, this includes: inputting the expression matrix and the first adjacency matrix into the region division module to obtain a first region division result; inputting the image patch into the feature extraction module for feature vector extraction to form an image feature expression matrix and a second adjacency matrix; inputting the image feature expression matrix and the second adjacency matrix into the region division module to obtain a second region division result; and using mutual information to perform image registration based on the first and second region division results to obtain the final region division result of the tissue slice.

[0045] Since stained image sections and sequencing tissue sections are usually derived from adjacent rather than the same tissue section, the spatial partitioning results of the image data S are used before adjusting and correcting the regional division boundaries of sequencing omics based on image information. i (i.e., the second region segmentation result, the moving image) was image-registered using mutual information to the spatial segmentation result S of the spatial transcriptomics data. m (i.e., the first region segmentation result, fixed image) in the same reference coordinate system. Specifically, the mutual information between two images can be reflected by entropy and joint entropy. Maximizing the mutual information between images through rigid transformations such as translation, rotation, and scaling is as follows:

[0046] MI(s i ,s m )=H(s i )+H(s m )-H(s i ,s m )

[0047] Where MI(s) i ,s m H(s) represents mutual information. i ) and H(s m H(s) represents the entropy of the moving image and the stationary image, respectively. i ,s m The joint entropy of the two images is denoted as . In the registered overlapping images, the spatial morphological information of cells can be used to adjust the spatial metabolomics domains to better reflect their actual distribution.

[0048] In an exemplary embodiment, the SpaMIC model constructed in this application further includes a feature fitting module; the feature fitting module is used to fit multimodal data. The SpaMIC model is also used to divide the tissue slices into regions based on the expression matrix, the first adjacency matrix, and other omics feature matrices. Specifically:

[0049] (1) Input the expression matrix, the first adjacency matrix and other omics feature matrices into the feature fitting module to obtain the integrated feature expression matrix.

[0050] The feature fitting module can perform fitting analysis on spatial metabolomics data and other omics data, combining features from two modalities. i (Representation matrix and first adjacency matrix) and y i (Other omics feature matrices) are input into a multimodal autoencoder. The encoder and decoder structures are the same as those in the SpaMIC_AE model, with the hidden layer dimension set to 1000-1000. The encoder outputs low-dimensional features z. x and z y An initial joint representation z is learned through a nonlinear neural network and used as the input feature of the decoder. Furthermore, low-dimensional features z... x and z y The feature representation h is computed separately via an additional neural network. x and h y The feature dimension is 128, and a contrastive learning strategy is used to adapt the data from the two modalities. The overall loss function includes a standard reconstruction loss based on the mean squared error of the two modalities and a contrastive learning loss function based on the normalized temperature-scaled cross-entropy loss function (NT-Xent loss). The multimodal autoencoder uses the Adam optimizer with an initial learning rate of lr = 0.0005 and training epochs of n = 500. The specific method is as follows:

[0051]

[0052]

[0053] z = f projector (concat(z x ,z y ))

[0054] h x =f projector (z x )

[0055] h y =f projector (z y )

[0056]

[0057] Among them, f encoderx (·) and f encodery (·) represents the decoder corresponding to the two modes in a multimodal autoencoder, f projector (·) represents a nonlinear neural network consisting of two linear layers and a ReLU activation layer, τ is the temperature parameter, sim(·) is the dot product of vectors, α is the weight of the reconstruction loss part (default is 1), and β is the weight of the contrastive loss part (default is 1).

[0058] (2) Input the integrated feature representation matrix and the first adjacency matrix into the region division module to obtain the region division result.

[0059] To further test the analytical capability of the SpaMIC model constructed in this application in the spatial structural domain, and to verify its performance in boundary recognition and multi-omics feature fitting, this application selected three sets of spatial metabolomics data as applications, including the pig embryo dataset, the mouse kidney dataset, and the mouse Parkinson's disease spatial multimodal analysis (SMA) dataset.

[0060] (I) Implementation Case 1

[0061] (1) Data collection

[0062] The porcine embryo dataset was captured using a Thermo LTQ instrument employing desorption electrospray ionization (DESI) to obtain information from cross-sectional sections of porcine fetuses. It contains 4959 pixels and 10200 mass spectrometry peaks, with a mass range of m / z 150–1000. The standardized dataset is accessible via the R package Cardinal.

[0063] (2) Data processing

[0064] The original spatial metabolomics data of the tissue slices were transformed into an expression matrix X and a first adjacency matrix A. Then, the region segmentation module of the SpaMIC model was used to segment the chemical features of the tissue regions, with its parameters set to default. Similarly, the collected dataset was benchmarked using Louvain clustering, SmartGate, and GraphST (version 1.1.1) in Scanpy, with all parameters set to default, and compared with the SpaMIC method.

[0065] (3) Results Analysis:

[0066] H&E stained sections (e.g.) Figure 2As shown in (a) of the diagram, the main morphology of the pig embryo can be clearly divided, but the identification of some regional edges and areas with unclear staining remains ambiguous. The collected spatial metabolomics data was processed using the SpaMIC region segmentation module, extracting 3994 regionally representative molecular features from the original 10200 feature molecules, reducing the data size by approximately 61%, and also obtaining a better regional feature contour map (e.g., ...). Figure 2 (as shown in (b)). The results of SpaMIC region segmentation were also compared with other methods, where Louvain clustering in Scanpy reconstructed the structure of the porcine embryo to some extent (e.g., Figure 2 As shown in (c) in the figure, it is greatly affected by data noise, and there are many error interferences at the boundaries and inside of different organs. Although the SmartGate and GraphST methods can identify several major organs in pig embryos, such as the brain, heart and liver (e.g., Figure 2 (as shown in (d) and (e) in the figure), but there are cases where the divided organizational regions are identified as too large compared to the actual organizations and the number of identified types is insufficient.

[0067] (II) Implementation Case 2

[0068] (1) Data collection

[0069] Mouse kidneys (7-month-old female C57) were cryopreserved at -80°C and then cut into two continuous sections with thicknesses of 20 μm and 10 μm, respectively, using a cryostat. A desorption / spray ionization (DESI) source (Prosolia, USA) coupled with a Synapt G2-Si mass spectrometer (Waters, UK) was used for mass spectrometry imaging of the thicker section at a step size of 100 × 100 μm (i.e., pixel resolution). Chromatographic grade methanol (Merck, German) was used in the DESI experiments at a flow rate of 1.5 μL / min. Positive ion mode was selected, with a DESI capillary voltage of 4.2 kV and a sampling cone voltage of 80 V. The nebulized nitrogen pressure was 0.5 MPa, and the ion transfer capillary temperature was 150°C. The mass range was 200–1000 m / z, and the mass resolution was 10000. Thin sections were stained with standard hematoxylin and eosin (H&E) and then digitally scanned at 20x magnification (i.e., 0.5 × 0.5 μm pixel resolution). The acquired mouse kidney dataset was obtained from the ProteomeXchangeConsortium database (accession number PXD038876) (data from public databases was used for tool testing).

[0070] (2) Data processing

[0071] The original spatial metabolomics data of the tissue slices were transformed into an expression matrix X and a first adjacency matrix A. Then, the region segmentation module of the SpaMIC model was used to segment the chemical features of the tissue regions, resulting in a rough region segmentation contour map. Next, the feature extraction module of the SpaMIC model was used to integrate the H&E-stained tissue region segmentation information into the aforementioned region segmentation contour map, refining the segmentation of the tissue structure regions.

[0072] (3) Results Analysis

[0073] H&E histological staining of mouse kidney sections allows for the approximate division of kidney tissue into several main regions: cortex, medulla, and pelvis. These three regions are clearly presented in the spatial structure analysis (MSI) map using metabolomics data alone. However, the allocation of some metabolite information is not ideal. For example, a portion of metabolites on the left side of the pelvic region (MSI metabolite cluster 2) exhibits serrated protrusions, which appear as blank areas in the H&E-stained tissue region. This is because environmental noise interferes with the spatial clustering results. Subsequently, the tissue region division information from the H&E-stained optical image (such as...) is further analyzed. Figure 3 (a) shown in the figure is applied to boundary recognition adjustment, so that the spatial region structure division results (as shown in the figure) are improved. Figure 3 As shown in (b) in the figure, it can be clearly corrected, and its spatial clustering results are closer to the true partition (e.g., Figure 3 As shown in (c) in the figure, this further illustrates the necessity of using radiomics data from tissue imaging for region segmentation, and also reflects the superiority of the method and strategy in this application. The SpaMIC model combines features from two modalities to obtain a tissue structure distribution map with refined segmentation results (such as...). Figure 3 As shown in (d) in the figure, the boundary between the pelvic region and the medullary region is more consistent with the actual situation.

[0074] (III) Implementation Case 3

[0075] (1) Data collection

[0076] H&E images from the Spatial Multimodal Analysis (SMA) dataset of mouse Parkinson's disease are available in the Mendeley Data database. Raw MALDI-MSI data in imzML format, as well as MS / MS spectra in mzML and vendor-specific baf formats collected from tissues, are available in the Figshare database. In situ sequencing data are available in Zenodo. Raw spatial transcriptomics sequencing data, microscopic images, and Spaceranger output files for mouse samples are available in the Figshare database.

[0077] (2) Data processing

[0078] The original tissue slice mass spectrometry imaging data were transformed into an expression matrix X and a first adjacency matrix A. Then, the region segmentation module of the SpaMIC model was used to segment the tissue regions based on their chemical characteristics, resulting in a rough region segmentation outline. Next, the feature extraction module of the SpaMIC model was used to integrate the H&E-stained tissue region segmentation information into the aforementioned region segmentation outline, refining the segmentation of tissue structural regions. Finally, the integrative analysis module of the SpaMIC model was used to perform fitting analysis on the spatial metabolomics data and spatial transcriptomics data.

[0079] (3) Results Analysis

[0080] To demonstrate the advantages of SpaMIC in multimodal joint analysis, an integrated analysis was performed on the SNc slice data (SMA dataset number V11T16-085_C1). For example... Figure 4 As shown, the regional division using spatial metabolomics data alone (e.g.) Figure 4 (as shown in (a)) and the region partitioning that integrates spatial metabolomics and spatial transcriptomics data (e.g.) Figure 4 Compared to (b) shown in the figure, it was found that some small regions in the spatial metabolome division were not identified, resulting in the loss of important information. The integrated data, combined with information from H&E staining images, can make the region division clearer (e.g., Figure 4 (As shown in (c)). Comparing the partitioning results of these three types of data, it was found that the addition of transcriptome data resulted in SpaMIC partitioning an additional specific region, and this specific region is related to the dopamine-enriched sites (such as...). Figure 4 As shown in (d), this is consistent with the fact that SpaMIC can obtain a more refined spatial structure domain after integrating and analyzing multi-omics data and then dividing the region, which demonstrates the advantages of the feature fitting module in SpaMIC.

[0081] Based on the same inventive concept, this application also provides a system for implementing the tissue region division method based on spatial metabolomics data as described above. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more tissue region division system embodiments based on spatial metabolomics data provided below can be found in the limitations of the tissue region division method based on spatial metabolomics data described above, and will not be repeated here.

[0082] In one exemplary embodiment, a tissue region partitioning system based on spatial metabolomics data is provided, comprising:

[0083] The transformation module is used to transform spatial metabolomics data of tissue slices into expression matrices and first adjacency matrices.

[0084] The image patch cropping module is used to crop H&E stained slide images into multiple image patches.

[0085] The model building and training module is used to build and train the SpaMIC model; the SpaMIC model includes a region segmentation module and a feature extraction module.

[0086] The region segmentation module is used to segment the tissue slice into regions based on the expression matrix, the first adjacency matrix, and the image patch using a trained SpaMIC model.

[0087] In one exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments. The computer device may be a server or a terminal, and its internal structure diagram may be as follows: Figure 5 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data to be processed. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When executed by the processor, the computer program implements a tissue region partitioning method based on spatial metabolomics data.

[0088] Those skilled in the art will understand that Figure 5 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0089] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0090] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0091] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0092] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0093] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchain. The processors involved in the embodiments provided in this application may be, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc.

[0094] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0095] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for tissue region division based on spatial metabolomics data, characterized in that, include: Spatial metabolomics data of tissue sections were converted into expression matrices and first adjacency matrices; The H&E stained section image was cropped into multiple image blocks; Construct and train a SpaMIC model; the SpaMIC model includes a region segmentation module and a feature extraction module; The region segmentation module uses a graph attention network autoencoder, and the feature extraction module uses a ResNet-50 network structure. Based on the expression matrix, the first adjacency matrix, and the image patch, the trained SpaMIC model is used to divide the tissue slice into regions; specifically, this includes: inputting the expression matrix and the first adjacency matrix into the region division module to obtain a first region division result; The image patch is input into the feature extraction module for feature vector extraction to obtain an image feature representation matrix and a second adjacency matrix; the image feature representation matrix and the second adjacency matrix are input into the region segmentation module to obtain a second region segmentation result; based on the first region segmentation result and the second region segmentation result, image registration is performed using mutual information to obtain the final region segmentation result of the tissue slice.

2. The tissue region division method based on spatial metabolomics data according to claim 1, characterized in that, The region partitioning module includes two iterative training processes; During the first iteration of training, the sample representation matrix and the first sample adjacency matrix are used as inputs to the region partitioning module, and the output is the feature molecule of the sample representation matrix. The sample expression matrix and the first sample adjacency matrix are obtained by converting spatial metabolomics data from sample tissue slices; The sample expression matrix is ​​updated based on the feature molecules; In the second iteration of training, the updated sample representation matrix and the first sample adjacency matrix are used as inputs to the region partitioning module.

3. The tissue region division method based on spatial metabolomics data according to claim 1, characterized in that, The SpaMIC model also includes a feature fitting module; The SpaMIC model is also used to divide the tissue slices into regions based on the expression matrix, the first adjacency matrix, and other omics feature matrices.

4. The tissue region division method based on spatial metabolomics data according to claim 3, characterized in that, The tissue slices are divided into regions based on the expression matrix, the first adjacency matrix, and other omics feature matrices, specifically including: The expression matrix, the first adjacency matrix, and other omics feature matrices are input into the feature fitting module to obtain the integrated feature expression matrix; The integrated feature representation matrix and the first adjacency matrix are input into the region partitioning module to obtain the region partitioning result.

5. A tissue region partitioning system based on spatial metabolomics data, characterized in that, include: The transformation module is used to convert spatial metabolomics data of tissue slices into expression matrices and first adjacency matrices. The image patch cropping module is used to crop H&E stained slide images into multiple image patches; The model building and training module is used to build and train the SpaMIC model; the SpaMIC model includes a region partitioning module and a feature extraction module; the region partitioning module adopts a graph attention network autoencoder, and the feature extraction module adopts a ResNet-50 network structure. The region segmentation module is used to segment the tissue slice into regions based on the expression matrix, the first adjacency matrix, and the image patch using a trained SpaMIC model; specifically, it includes: inputting the expression matrix and the first adjacency matrix into the region segmentation module to obtain a first region segmentation result; The image patch is input into the feature extraction module for feature vector extraction to obtain an image feature representation matrix and a second adjacency matrix; the image feature representation matrix and the second adjacency matrix are input into the region segmentation module to obtain a second region segmentation result; based on the first region segmentation result and the second region segmentation result, image registration is performed using mutual information to obtain the final region segmentation result of the tissue slice.

6. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the tissue region partitioning method based on spatial metabolomics data as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the tissue region division method based on spatial metabolomics data as described in any one of claims 1-4.

Images