High-throughput single cell space transcriptome analysis method based on multiple rounds of imaging coding

Abstract

The invention relates to a high-throughput single cell space transcriptome analysis method based on multiple rounds of imaging coding. The invention provides a high-throughput single cell space transcriptome analysis method based on multi-round imaging coding, which comprises the following steps of: preprocessing an image to be processed through a low-pass filter designed according to a microscope optical system, obtaining a displacement correction affine matrix of a multi-round imaging image through a registration algorithm according to a dyed image in the preprocessed image, and obtaining a displacement correction affine matrix of the multi-round imaging image; shimming correction, background noise removal, image enhancement and enhancement of signal points in the image are performed on each round and each channel image of the to-be-processed image to obtain an enhanced to-be-processed image, a local extreme value is obtained through a Gaussian Laplacian operator and morphological operation, the extreme value is selected as a signal point coordinate when the numerical value of the extreme value is higher than a preset threshold, and the signal point coordinate is selected as a signal point coordinate when the numerical value of the extreme value is lower than the preset threshold. And obtaining a cell segmentation result, distributing signal points to the determined single cells, and converting the obtained coordinates of the signal points into a final single cell space transcriptome expression quantity matrix through a decoding algorithm and a coding table.

Description

Technical Field

[0001] The present invention relates to an analysis process of a high-throughput RNA in situ hybridization detection method realized by multiple rounds of imaging combined encoding. Background Art

[0002] Single-cell spatial transcriptomics is an emerging field that provides a new approach to understanding the spatial distribution of gene expression within tissues. In traditional single-cell transcriptomics, spatial information about gene expression in tissues is lost during the extraction process. Therefore, while traditional single-cell transcriptomics can provide a lot of information about the genes being expressed in a sample, it cannot provide information about the location of gene expression in the tissue. Single-cell spatial transcriptomics addresses this limitation by retaining spatial information during RNA extraction (or in situ capture). This allows researchers to determine not only which genes are being expressed, but also where they are expressed in the tissue. This additional spatial information can provide valuable information for understanding the development and function of organs and tissues, as well as the progression of diseases such as cancer and response to treatment.

[0003] Spatial transcriptome technology can be divided into two categories: one is based on probe hybridization and imaging, such as smFISH, MERFISH, etc.; the other is based on sequencing, including 10x VISIUM, DBiT-seq, Stereo-seq, etc. However, for spatial transcriptome images, the current analysis process, such as seqFISH, has the following limitations: 1) The image processing step requires a background image to achieve a forward-looking image correction. In the actual implementation, it is often difficult to obtain a matching background image due to reasons such as machine displacement stability; 2) The threshold selection of signal points in the process depends on manual judgment, which is relatively subjective and heavily dependent on manual labor and cannot be automated; 3) The decoding algorithm is inefficient and slow, and there is a problem that it may not be accurately resolved when the signal points are too dense; 4) The decoding algorithm needs to provide cell segmentation results first, and the decoding is limited to a single cell. If the cell segmentation is optimized later, the decoding process must be repeated, which wastes computing power and time; 5) The overall process is written in matlab, with insufficient processing efficiency and speed, and because it is based on the matlab platform, the cost is high.

[0004] Therefore, for methods based on probe hybridization and imaging, there is currently a lack of efficient and automatic upstream image analysis and decoding methods. This field is in urgent need of a method that can accurately and efficiently achieve in situ analysis of RNA molecules. Summary of the invention

[0005] The present invention aims to solve the problem of lack of an efficient, accurate and automated method for analyzing RNA molecule coordinates.

[0006] In some aspects, provided herein are inventions described in one or more of the following.

[0007] A high-throughput single-cell spatial transcriptome analysis method based on multiple rounds of imaging encoding comprises the following steps:

[0008] Step S10, preprocessing the image to be processed according to a low-pass filter designed for the microscope optical system;

[0009] Step S20, according to the staining image (such as the cell nucleus staining image, such as the DAPI image) in the preprocessed image, a displacement-corrected affine matrix of the multi-round imaging images is obtained by a registration algorithm.

[0010] Step S30, performing shimming correction, background noise removal, image enhancement, and strengthening signal points in the image on each round and each channel image of the image to be processed to obtain an enhanced image to be processed;

[0011] Step S40, obtaining local extrema through Gaussian Laplacian (LoG) operator and morphological operation, when the extremum value is higher than the preset threshold, it is selected as the coordinate of the signal point, preferably by a fast radial symmetric center point determination algorithm, optimizing the coordinates of the signal point, and realizing sub-pixel precise positioning, preferably using the affine matrix obtained in step S20 to correct the signal points determined by a single round of images, so that the signal points obtained by multiple rounds of images are converted to the same coordinate system, preferably determining the dispersion correction matrix between multiple channels through fluorescent beads, and applying it to the coordinates of the signal points obtained in the previous step, so that the signal points obtained by multiple rounds of multi-channel images are in the same coordinate system;

[0012] Step S50, obtaining the result of cell segmentation and assigning the signal points to the determined single cells;

[0013] Step S60, converting the obtained signal point coordinates into the final single-cell spatial transcriptome expression matrix through a decoding algorithm and a pre-designed coding table.

[0014] In some preferred embodiments, step S10 includes:

[0015] Step S11, obtain a fluorescent bead image with a size similar to that of the signal point, transform the spatial domain image into the frequency domain through Fourier transform, and move the zero-frequency component to the center of the spectrum.

[0016] Step S12, a low-pass filter is designed according to the frequency domain to retain the real signal, and its formula is expressed as:

[0017]

[0018] Among them, x and y represent the x and y coordinates in the frequency domain, cut represents the frequency filtering range, and exp represents the order of the filter. It is preferred to adjust cut and exp according to the real bead frequency domain signal so as to retain the signal and filter the noise as much as possible.

[0019] Step S13, performing an inverse zero-frequency shift on the frequency domain, and then converting the image from the frequency domain back to the spatial domain by inverse Fourier transform to obtain a corrected image.

[0020] In some preferred embodiments, step S20 includes:

[0021] Step S21, extracting image features through SIFT (Scale-Invariant Feature Transform) algorithm.

[0022] Step S22, obtaining the optimal affine matrix that matches the features of the two images by iterating the RANSAC (Random Sample Consensus) algorithm.

[0023] In some preferred embodiments, step S30 includes:

[0024] Step S31, shimming correction of the image to be processed, using the tophat algorithm (using a disk structure element with a radius of 30), to process the image to be processed to remove the effects of uneven light and tissue background,

[0025] Step S32, background denoising of the image to be processed after shimming correction, by processing the image obtained in step S31 using ROF (Rudin-Osher-Fatemidenoising) to obtain a corrected image.

[0026] Step S33, image enhancement: An image of fluorescent beads of similar physical size to the signal size is acquired, and the point spread function (PSF) of the microscope system is estimated using the image. The corrected image obtained in step S32 is enhanced using the Lucy-Richardson deconvolution algorithm and the estimated PSF.

[0027] In some preferred embodiments, step S40 includes:

[0028] Step S41, automatic threshold determination, setting the threshold search space, and performing the following operations for each threshold to be searched: using a rotationally symmetric Gaussian Laplacian filter with a size of h and a standard deviation of s on the enhanced image obtained in S33 to obtain image I1, and comparing the image with the threshold to be searched, resetting it to 0 if it is below the threshold to obtain image I2, and using the structural element The image I3 is obtained by dilating I2. The position where I2 is greater than I3 is determined as the local maximum point (the proposed signal). By determining the inflection point of the curve of the threshold sequence and the number of signal points corresponding to the threshold, the inflection point is defined as the proposed automatic threshold. This point represents the dividing line between the potential background and the real signal.

[0029] Step S42, by applying a rotationally symmetric Gaussian Laplacian filter with a size of h and a standard deviation of s to the enhanced image obtained in S33, obtaining an image I1, and comparing the image with the automatic threshold obtained in S41, resetting the image below the threshold to 0 to obtain an image I2, and using the structural element Perform a dilation morphological operation on I2 to obtain image I3. The position where I2 is greater than I3 is determined as the local maximum point (the proposed signal).

[0030] Step S43: for each signal point obtained in S42, obtain a 3x3 image around it, and obtain the sub-pixel position estimation of the point through a fast radial symmetric center point determination algorithm.

[0031] Step S44, using the inter-round displacement correction affine matrix and the inter-channel dispersion correction affine matrix to perform displacement correction on the points obtained in S43.

[0032] In some preferred embodiments, step S60 includes:

[0033] Step S61, for each signal point in the cell, a graph G(V,E) is constructed, where the nodes are signal points (V) and the edges (E) are the affinity between the signal points. Preferably, by cutting the graph, some nodes with a degree less than n-1 (n is the minimum code recognized as valid decoding) are pre-cleaned, and the edges connecting the two sides belonging to the same hybridization round and the same hybridization channel are cleaned. By cutting the nodes that do not meet the decoding rules, the amount of calculation is reduced. By traversing each V i As a seed, search for all other V within the preset radius r. j Construct a subgraph, match candidate codes within the subgraph, annotate the candidate codes to genes according to the coding table, and calculate the average distance between the points of the coding combination as the loss metric.

[0034] Step S62, for all potential candidate combinations, filter the decoding combinations with less than n nodes in the coding subgraph (n is the minimum code recognized as a valid decoding), ensure that each coding subgraph has at least n different fluorescent signals, ensure the error correction capability, sort according to the number of nodes and loss metric (average Euclidean distance between nodes in the subgraph) of each potential decoding subgraph, and add the coding subgraphs with the most signal point matches (most confident) and the smallest loss (smallest Euclidean distance between each other) to the final coding result in sequence. If a signal point in the later coding has already appeared in an earlier coding, then discard the signal point to ensure that each signal point is only used once.

[0035] Through S61 and S62, the following principles are guaranteed: an RNA molecule must meet the following conditions to be resolved: (1) at least three fluorescent spots are detected. Since the Hamming distance of the coding table is 1, it can be guaranteed that even if one fluorescent spot is missing due to accidental circumstances, it can be accurately resolved; (2) the average distance of a set of 4 or 3 fluorescent spots encoded is smaller than that of other combinations, in order to satisfy a basic fact that the distance between the four fluorescent spots from the same RNA molecule must be smaller than that of the fluorescent spots of other RNA molecules due to the steric hindrance between RNA molecules.

[0036] In some embodiments, the present invention provides a computer system comprising:

[0037] processor;

[0038] A memory having instructions stored therein, wherein when the instructions are executed by the processor, the processor is caused to execute the method described in any one of the embodiments of the present invention.

[0039] In some embodiments, the present invention provides a computer-readable storage medium storing instructions, which, when executed by a processor, causes the processor to perform a method described in any one of the embodiments of the present invention.

[0040] In some embodiments, the present invention provides a computer program product comprising instructions, which, when executed by a processor, cause the processor to perform the method described in any one of the embodiments of the present invention.

[0041] The beneficial effects of the present invention are in one or more of the following aspects:

[0042] (1) The present invention processes the image through an image processing algorithm, and does not need to obtain a background image that matches the experimental image, thereby simplifying the requirements for the experimental process and the stability of the experimental equipment.

[0043] (2) The present invention implements an automatic threshold search scheme, which realizes automated threshold setting for multiple fields of view, multiple hybridization rounds, and multiple channels. Compared with manual setting, it saves manpower and is more objective.

[0044] (3) The decoding algorithm of the present invention adopts the idea of ​​graph theory, and regards the decoding problem as the segmentation of subgraphs. By cutting the graph, removing the edges that do not meet the preconditions, simplifying the graph, and reducing the amount of computation, the present invention avoids the low running efficiency of other algorithms due to the inclusion of a large number of loop structures. In an actual test (with a sample size of 158), the algorithm of the present invention achieved an efficient decoding speed of 225.68±131.37 seconds, which is better than the running speed of 4012.09±1567.94 seconds of other algorithms. The decoding algorithm of the present invention has super high decoding performance, achieving nearly 18 times the performance.

[0045] (4) The decoding algorithm of the present invention does not depend on the cell segmentation results. It can first distribute the signal points to each cell according to the cell segmentation results to complete the decoding, or it can directly decode a single shooting field of view and then assign the decoded RNA molecules to the corresponding cells according to their coordinates. Other algorithms depend on the segmentation results and can only segment first and then decode.

[0046] (5) The algorithm of the present invention can be based on free Python, and can achieve a much higher computing speed than similar algorithms through GPU acceleration. Through GPU acceleration, in actual tests (sample size 158), the image processing part of this process achieved a score of 121.74±40.28 seconds, while the image processing part of other processes only took 916.20±197.91 seconds, achieving a speed increase of nearly 8 times. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the process of the single-cell spatial transcriptome analysis method based on multiple rounds of fluorescent probe hybridization imaging of the present invention.

[0048] Figure 2 This is an example diagram of the multi-round imaging registration results of the single-cell spatial transcriptome analysis method based on multi-round fluorescent probe hybridization imaging of the present invention.

[0049] Figure 3 This is an example diagram of processed images of the single-cell spatial transcriptome analysis method based on multiple rounds of fluorescent probe hybridization imaging of the present invention.

[0050] Figure 4 This is an example diagram of the automatic threshold detection results of the single-cell spatial transcriptome analysis method based on multiple rounds of fluorescent probe hybridization imaging of the present invention.

[0051] Figure 5 This is an example diagram of detection signal points of the single-cell spatial transcriptome analysis method based on multiple rounds of fluorescent probe hybridization imaging of the present invention.

[0052] Figure 6This is an example diagram of the decoding results of the single-cell spatial transcriptome analysis method based on multiple rounds of fluorescent probe hybridization imaging of the present invention. DETAILED DESCRIPTION

[0053] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0054] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0055] The present invention provides a single-cell spatial transcriptome method based on multi-round imaging encoding, the method comprising:

[0056] Step S10, preprocessing the image to be processed according to a low-pass filter designed for the microscope optical system;

[0057] Step S20, obtaining the displacement-corrected affine matrix of multiple rounds of imaging images through a registration algorithm according to the DAPI-stained image in the preprocessed image;

[0058] Step S30, performing shimming correction, background noise removal, image enhancement, and strengthening signal points in the image on each round and each channel image of the image to be processed to obtain an enhanced image to be processed;

[0059] Step S40, obtain local extrema through Gaussian Laplacian operator (LoG) and morphological operations, and the extrema value is selected as the signal point coordinate when it is higher than the preset threshold. Optimize the signal point coordinates through a fast radially symmetric center point determination algorithm to achieve sub-pixel precise positioning. Use the affine matrix obtained in step S20 to correct the signal points determined by a single round of images, so that the signal points obtained by multiple rounds of images are converted to the same coordinate system. Determine the dispersion correction matrix between multiple channels through fluorescent beads, and apply it to the signal point coordinates obtained in the previous step, so that the signal points obtained from multiple rounds of multi-channel images are in the same coordinate system;

[0060] Step S50, obtaining the result of cell segmentation and assigning the signal points to the determined single cells;

[0061] Step S60, converting the obtained signal point coordinates into the final single-cell spatial transcriptome expression matrix through a decoding algorithm and a pre-designed coding table.

[0062] The analysis process of the high-throughput RNA in situ hybridization detection method based on multi-round imaging combined encoding in the first embodiment of the present invention includes steps S10 to S60, and each step is described in detail as follows:

[0063] Step S10, preprocessing the image to be processed using a low-pass filter designed according to the optical system of the microscope;

[0064] Step S11, using the same optical system, obtain an image of fluorescent beads with a size similar to that of the signal spot, convert the original image from the spatial domain to the frequency domain through Fourier transform, and move the zero-frequency component to the center of the spectrum.

[0065] Step S12, according to the signal condition in the frequency domain, a specific low-pass filter is designed to filter out high-frequency noise and retain the real signal.

[0066] Step S13, performing an inverse zero-frequency shift on the filtered frequency domain signal, and then converting the image from the frequency domain back to the spatial domain by inverse Fourier transform to obtain a corrected image.

[0067] Step S20, obtaining the displacement-corrected affine matrix of multiple rounds of imaging from the DAPI-stained image of the preprocessed image through a registration algorithm.

[0068] Step S21, extracting image features by SIFT algorithm, and obtaining paired sample points of images of multiple rounds of imaging relative to the reference round image, for subsequent estimation of the affine matrix.

[0069] Step S22, using the RANSAC algorithm to iteratively and randomly select paired sample points to improve the robustness of the estimated affine matrix and obtain the optimal affine matrix. Figure 2 As shown in FIG. 1 , the registration result of the two images after applying the affine matrix is ​​obtained. The reference wheel is green, the wheel to be registered is magenta, and the two colors are added to form gray after the two are matched.

[0070] Step S30, performing operations such as non-uniform light correction, background noise removal, image enhancement and signal point strengthening on each hybridization round and each channel of the image to be processed to obtain an enhanced image for signal point detection.

[0071] Step S31, performing uneven light correction on the image to be processed. The image to be processed is processed by using the tophat algorithm (using a disk structure element with a radius of 30) to remove the effects of uneven light and tissue background.

[0072] Step S32, background denoising of the image to be processed after shimming correction. The image obtained in step S31 is processed using ROF (Rudin-Osher-Fatemidenoising) to obtain a corrected image, remove background noise, and further enhance the image.

[0073] Step S33, image enhancement. Obtain an image of fluorescent beads of similar physical size to the signal size, and use this image to estimate the point spread function (PSF) of the microscope system. Use the Lucy-Richardson deconvolution algorithm and the estimated PSF to enhance the corrected image obtained in step S32, so that the signal is further highlighted and the background is weakened. Figure 3 As shown, the final image enhancement result is compared with the original image. It can be seen that the signal is significantly enhanced compared to the background through processing, which is conducive to subsequent analysis.

[0074] Step S40, obtain local extrema through Gaussian Laplace operator (LoG) and morphological operation, and the extremum value is selected as the signal point coordinate when it is higher than the preset threshold. The signal point coordinates are optimized through a fast radially symmetric center point determination algorithm to achieve sub-pixel precise positioning. The signal points determined by a single round of images are corrected using the affine matrix obtained in step S20, so that the signal points obtained by multiple rounds of images are converted to the same coordinate system. The dispersion correction matrix between multiple channels is determined by fluorescent beads, and it is applied to the signal point coordinates obtained in the previous step, so that the signal points obtained by multiple rounds of multi-channel images are in the same coordinate system.

[0075] Step S41, by running the signal point detection algorithm for each assumed threshold in the preset automatic threshold exploration space, the number of signal points under the threshold state is obtained, and then the relationship between the number of signal points and the threshold is plotted as shown in Figure 4 The curve is obtained by fitting a straight line on both sides of the assumed threshold and calculating the mean square error between the straight line and the curve. When the fitting error is the smallest, the point is determined as the final threshold, representing the dividing line between the real signal and the potential background. Figure 4 , which is an example of automatic threshold determination.

[0076] Step S42, filtering the enhanced image obtained in S33 by using a Gaussian Laplacian filter, and comparing the filtered image with the threshold determined in S41, resetting the part below the threshold to 0, and then using a 3x3 structure element with a center of 0 and all other positions being 1 to perform morphological operations on the reset image, comparing the values ​​of the images before and after the morphological operation, the point before the operation that is larger than the point after the operation is the local maximum, which is determined as the coordinate of the signal point.

[0077] Step S43, for each signal point obtained in S42, obtain the corrected image obtained in S32 within a 3x3 range centered on the signal point, and estimate the sub-pixel position of the point through a radially symmetric center point algorithm.

[0078] Step S44, applying the multi-hybridization inter-round displacement correction matrix obtained in step S22 and the pre-acquired multi-channel simple dispersion correction matrix to the coordinate points obtained in step S43, to obtain the corrected final coordinates for subsequent operations.

[0079] Step S50: The cell segmentation result is represented as a closed polygon formed by connecting the cell boundary coordinates. The signal point is assigned to a single cell by determining whether the signal point falls within the polygon. Figure 5 This is the final result of signal point and cell segmentation. The red dots mark the locations of signal points that are higher than the automatic threshold.

[0080] Step S60, converting the obtained signal point coordinates into the final single-cell spatial transcriptome expression matrix through a decoding algorithm and a pre-designed coding table.

[0081] Step S61: For each signal point in each cell, a graph G(V,E) is constructed, where the nodes are signal points (V) and the edges (E) are the affinity between signal points. i As a seed, search for all other V within the preset radius r. j A subgraph is constructed, candidate encodings are matched within the subgraph, and the candidate encodings are annotated to genes according to the encoding table. At the same time, the average distance between the points of the encoding combination is calculated as the loss metric.

[0082] Step S62, for all potential candidate combinations, filter the combinations with less than 3 seeds in the encoding to ensure that each successfully encoded molecule has at least 3 different fluorescent signals to improve the error correction capability. Sort by the number of seed matches and the loss metric, and add the encodings with the most seed matches (most confident) and the smallest loss (smallest Euclidean distance between each other) to the final encoding result in order. If a signal point in the later encoding has already appeared in an earlier encoding, discard the signal point to ensure that each signal point is only used once. Finally, the final parsing result is obtained, that is, an expression matrix with known coordinates of each RNA molecule. Figure 6 , which is an example of a partial parsing result.

[0083] It should be noted that the system provided in the embodiment of the present application is only illustrated by the division of the above-mentioned functional modules. In practical applications, the above-mentioned function allocation can be completed by different functional modules as needed, that is, the modules or steps in the embodiment of the present invention can be decomposed or combined again. For example, the modules in the above-mentioned embodiment can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiment of the present invention are only for distinguishing each module or step, and are not regarded as improper limitations of the present invention.

[0084] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A high-throughput single-cell spatial transcriptome analysis method based on multiple rounds of imaging encoding, comprising the following steps: Step S10, preprocessing the image to be processed by a low-pass filter designed according to the microscope optical system, Step S20, obtaining the displacement-corrected affine matrix of multiple rounds of imaging images through a registration algorithm according to the stained image in the preprocessed image, Step S30, performing shimming correction, background noise removal, image enhancement, and strengthening signal points in the image for each round and each channel image to be processed to obtain an enhanced image to be processed. Step S40, obtaining local extreme values ​​through Laplacian of Gaussian (LoG) and morphological operations, and when the extreme value is higher than a preset threshold, it is selected as the coordinate of the signal point. Step S50, obtaining the result of cell segmentation and assigning the signal points to the determined single cells, Step S60, converting the obtained signal point coordinates into the final single-cell spatial transcriptome expression matrix through a decoding algorithm and a coding table.

2. The method of claim 1, wherein step S10 comprises: Step S11, obtain a fluorescent bead image with a size similar to that of the signal point, transform the spatial domain image into the frequency domain through Fourier transform, and move the zero-frequency component to the center of the spectrum. Step S12, design a low-pass filter according to the frequency domain conditions to filter out high-frequency noise and retain the real signal. Step S13, performing an inverse zero-frequency shift on the frequency domain, and then converting the image from the frequency domain back to the spatial domain by inverse Fourier transform to obtain a corrected image.

3. The method of claim 1 or 2, wherein step S20 comprises: Step S21, extracting image features through SIFT algorithm, obtaining paired sample points of multiple rounds of imaging images relative to the reference round image, for subsequent estimation of the affine matrix, Step S22, obtaining the optimal affine matrix that matches the features of the two images through iteration of the RANSAC algorithm.

4. The method according to any one of claims 1 to 3, wherein step S30 comprises: Step S31, shimming correction of the image to be processed, wherein the image to be processed is processed by using the tophat algorithm to remove the influence of uneven light and tissue background. Step S32, background denoising of the image to be processed after shimming correction, wherein the image obtained in step S31 is processed using ROF to obtain a corrected image. Step S33, image enhancement, wherein an image of fluorescent beads of similar physical size to the signal size is acquired, the point spread function PSF of the microscope system is estimated using the image, and the corrected image obtained in step S32 is enhanced using the Lucy-Richardson deconvolution algorithm and the estimated PSF.

5. The method according to any one of claims 1 to 4, wherein step S40 comprises: Step S41, by running the signal point detection algorithm for each assumed threshold value of the preset automatic threshold exploration space, the number of signal points under the threshold state is obtained, and then a curve is drawn based on the relationship between the number of signal points and the threshold value, and a straight line is fitted on both sides of the assumed threshold value, and the fitting mean square error between the straight line and the curve is calculated. When the fitting error is the smallest, the point is determined as the final threshold value, which represents the dividing line between the real signal and the potential background. Step S42, filtering the enhanced image obtained in S33 with a Gaussian Laplacian filter, and comparing the filtered image with the threshold determined in S41, resetting the part below the threshold to 0, and then performing a morphological operation on the reset image using a 3x3 structure element with the center at 0 and all other positions at 1, comparing the values ​​of the image before and after the morphological operation, the point before the operation that is larger than the point after the operation is the local maximum, that is, the coordinates of the signal point. Step S43, for each signal point obtained in S42, obtain the corrected image obtained in S32 within a 3x3 range centered on the signal point, and estimate the sub-pixel position of the point through a radially symmetric center point algorithm. Step S44, using the inter-round displacement correction affine matrix and the inter-channel dispersion correction affine matrix to perform displacement correction on the points obtained in S43.

6. The method according to any one of claims 1 to 5, wherein step S60 comprises: Step S61: for each signal point in the cell, a graph G(V,E) is constructed, where the nodes are signal points (V) and the edges (E) are the affinity between signal points. t As a seed, search for all other V within the preset radius r. j Construct a subgraph, match candidate codes within the subgraph, annotate the candidate codes to genes according to the coding table, and calculate the average distance between the points of the coding combination as the loss metric. Step S62, for all potential candidate combinations, filter the decoding combinations with less than n nodes in the coding subgraph, where n is the minimum code recognized as a valid decoding, ensure that each coding subgraph has at least n different fluorescent signals, ensure error correction capability, sort according to the number of nodes and loss metric of each potential decoding subgraph, and add the coding subgraphs with the most signal point matches, i.e. the most confident and the smallest loss, i.e. the smallest Euclidean distance between each other, to the final coding result in sequence. If a signal point in the later coding has already appeared in an earlier coding, then discard the signal point to ensure that each signal point is used only once.

7. A computer system comprising: processor; A memory having instructions stored thereon, which, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 6.

8. A computer-readable storage medium storing instructions, which, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 6.

Images