A method, medium and system for outputting cell removal signal by flow cytometer

Abstract

The present invention provides a method, medium and system for outputting a cell cluster removal signal by a flow cytometer, belonging to the technical field of flow cytometers, the method comprising: traversing each electric pulse signal in an FSC pulse set, obtaining an electric pulse signal whose channel value is greater than a first threshold, and forming an FSC large pulse set; screening an FSC matrix, an SSC pulse matrix and a fluorescent pulse matrix according to the FSC large pulse set; establishing a polymer matrix with the obtained screened FSC matrix, the screened SSC pulse matrix and the screened fluorescent pulse matrix; analyzing the polymer matrix using a cell cluster analysis model, obtaining the electric pulse signal of the corresponding cell cluster in the polymer matrix; deleting the electric pulse signal of the corresponding cell cluster in the electric pulse signal set. The present invention can solve the technical problem that the existing cell cluster identification and removal methods rely on manually set thresholds and parameters, require manual adjustment of identification rules for different samples, and cannot be automated.

Description

Technical Field

[0001] The present invention belongs to the technical field of flow cytometers, and in particular, relates to a method, a medium and a system for outputting a cell removal signal by a flow cytometer. Background Art

[0002] Flow cytometry is a rapid, high-throughput single-cell analysis technique that is widely used in basic research and clinical testing. In flow cytometry measurements, individual cells are arranged in thin lines through the fluid, passing through the field of view where the laser and the detector intersect one by one, and various optical signals are recorded. Based on the forward scattered light (FSC), side scattered light (SSC) and fluorescence signals, the size, endoplasmic and surface marker characteristics of each cell can be analyzed.

[0003] Compared with single cells, cell clusters cause serious interference to the quantitative analysis of flow cytometry. Cell clusters are formed by the aggregation of multiple cells, which are larger in size and produce stronger optical signals than single cells. This will seriously affect the subsequent quantitative analysis and cell classification. Therefore, identifying and removing cell clusters is an important preprocessing step for flow cytometry data analysis.

[0004] There are two main methods for identifying cell clusters:

[0005] (1) Pulse shape-based methods. The pulse waveform information generated when each cell or cell cluster passes through the detection area, such as pulse area, width, peak ratio, etc., is used to set a threshold for identification. This type of method relies on parameter settings and empirical judgment.

[0006] (2) Snapshot image-based methods. First, a tiny image of each event is captured, and then the cell clusters are identified through image processing technology. However, this method requires the storage of a large number of images, is computationally complex, and is costly and not widely applicable.

[0007] In general, existing cell cluster identification and removal methods rely on manually set thresholds and parameters, and require manual adjustment of identification rules for different samples, which cannot be automated. Summary of the invention

[0008] In view of this, the present invention provides a method, medium and system for outputting a cell cluster removal signal by a flow cytometer, which can solve the technical problem that the existing cell cluster identification and removal methods rely on manually set thresholds and parameters, require manual adjustment of identification rules for different samples, and cannot be automated.

[0009] The present invention is achieved in that:

[0010] A first aspect of the present invention provides a method for outputting a cell removal signal by a flow cytometer, comprising the following steps:

[0011] S10, collecting a set of electric pulse signals collected by the photoelectric converter of the flow cytometer, including a set of electric pulse signals generated by forward scattered light, a set of electric pulse signals generated by side scattered light, and a set of electric pulse signals generated by fluorescence signals, which are respectively recorded as FSC pulse set, SSC pulse set, and fluorescence pulse set; wherein the electric pulse signal includes time, length, width, and area;

[0012] S20, establishing an FSC matrix, an SSC pulse matrix, and a fluorescence pulse matrix respectively using the obtained FSC pulse set, SSC pulse set, and fluorescence pulse set according to the electrical pulse signal;

[0013] S30, traversing each electrical pulse signal in the FSC pulse set, acquiring electrical pulse signals whose channel values ​​are greater than a first threshold, and forming an FSC large pulse set;

[0014] S40, screening the FSC matrix, the SSC pulse matrix and the fluorescence pulse matrix according to the FSC large pulse set to obtain a screening FSC matrix, a screening SSC pulse matrix and a screening fluorescence pulse matrix;

[0015] S50, establishing a polymerization matrix using the obtained screening FSC matrix, screening SSC pulse matrix, and screening fluorescence pulse matrix;

[0016] S60, analyzing the aggregation matrix using a pre-trained cell cluster analysis model to obtain electrical pulse signals corresponding to the cell clusters in the aggregation matrix;

[0017] S70, deleting the electrical pulse signal corresponding to the cell cluster from the electrical pulse signal set, and outputting the obtained electrical pulse signal set to the data processing system of the flow cytometer.

[0018] The technical effect of a method for outputting a cell cluster removal signal by a flow cytometer provided by the present invention is as follows: since there is a certain connection between the FSC, SSC, and fluorescence signal of each cell or cell cluster, for example, the cell cluster is relatively large in size and has a more complex internal structure, FSC and SSC will have a certain correlation. At the same time, the fluorescence signal of the cell cluster will have a different channel value from that of ordinary cells due to the relatively large number of fluorescent dyes in the cell cluster. Simple manual judgment will not be accurate enough due to the large amount of data. Therefore, the multi-channel fusion method is used to perform multi-feature extraction and analysis using an aggregation matrix, so that the cell cluster signal can be better determined.

[0019] On the basis of the above technical solution, the method for outputting a cell removal signal by a flow cytometer of the present invention can also be improved as follows:

[0020] Among them, the step of establishing FSC matrix, SSC pulse matrix and fluorescence pulse matrix respectively with the obtained FSC pulse set, SSC pulse set and fluorescence pulse set according to the electrical pulse signal also includes the step of eliminating noise points of the electrical pulse signal.

[0021] The step of traversing each electrical pulse signal in the FSC pulse set, acquiring the electrical pulse signal whose channel value is greater than the first threshold, and forming the FSC large pulse set specifically includes:

[0022] Traversing the FSC pulse set, extracting the pulse signal whose FSC channel value is greater than the first threshold as the first pulse set;

[0023] Calculate the inter-pulse distance of the first pulse set and merge adjacent pulses;

[0024] Performing smoothing filtering on the first pulse set after merging adjacent pulses;

[0025] The PCA method is used to reduce the dimension of the first pulse set after smoothing filtering.

[0026] The first pulse set after dimension reduction is used as the FSC large pulse set.

[0027] By default, the first threshold is set to 3 times the FSC baseline noise.

[0028] The steps of screening the FSC matrix, the SSC pulse matrix and the fluorescence pulse matrix according to the FSC large pulse set to obtain the steps of screening the FSC matrix, screening the SSC pulse matrix and screening the fluorescence pulse matrix specifically include:

[0029] Traverse the FSC matrix and keep the rows corresponding to the large FSC pulses;

[0030] According to the large FSC pulse, the corresponding row was screened from the SSC and fluorescence matrix;

[0031] The matrix after smoothing filter screening;

[0032] Output the smoothed FSC, SSC and fluorescence matrix.

[0033] The step of establishing a polymerization matrix using the obtained screening FSC matrix, the screening SSC pulse matrix and the screening fluorescence pulse matrix specifically includes:

[0034] The screening smoothed matrices of different channels were horizontally spliced ​​to form an aggregate matrix;

[0035] Normalize the aggregation matrix;

[0036] The normalized aggregation matrix is ​​subjected to dimensionality reduction using PCA to obtain a dimensionality-reduced aggregation matrix that integrates information from different channels.

[0037] The steps of establishing and training the cell cluster analysis model specifically include:

[0038] Collect multiple historical aggregation matrix samples and use them as training samples after labeling the cell cluster data;

[0039] Build machine learning classification models;

[0040] The training samples are used to train the machine learning classification model to obtain a cell cluster analysis model.

[0041] Furthermore, the model used to build the machine learning classification model is SVM or random forest.

[0042] The step of deleting the electric pulse signal corresponding to the cell cluster from the electric pulse signal set specifically includes:

[0043] Traversing the original electrical pulse signal set;

[0044] Determine whether each pulse is in the identified cell cluster set;

[0045] If so, delete the pulse from the original set;

[0046] The pulses of non-cell clusters are retained to form a filtered output set;

[0047] Returns the filtered electrical pulse signal set.

[0048] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for outputting a decellularization signal by a flow cytometer.

[0049] A third aspect of the present invention provides a system for outputting a decellularization signal by a flow cytometer, which includes the above-mentioned computer-readable storage medium.

[0050] Compared with the prior art, the method, medium and system for outputting a cell removal signal by a flow cytometer provided by the present invention have the following beneficial effects:

[0051] 1. Automatic identification and removal of cell clusters

[0052] The present invention can automatically identify cell clusters in the aggregation matrix by constructing a pre-trained machine learning model without manually setting identification rules and thresholds. This avoids the complexity of parameter adjustment and realizes intelligent automatic identification of cell clusters.

[0053] In contrast, the existing threshold-based methods require professionals to adjust parameters for different samples, are highly dependent on experience, and cannot achieve automatic batch processing. The automatic identification method of the present invention greatly simplifies the operation process.

[0054] 2. Improved the accuracy of cell cluster identification

[0055] The present invention comprehensively utilizes multi-channel information such as FSC and SSC to construct a polymerization matrix with strong expression ability, and adopts a pre-trained machine learning model to ensure the accuracy of cell cluster identification.

[0056] Existing methods often cause missed recognition or misrecognition due to improper threshold setting. However, the present invention provides richer features through a multi-channel aggregation matrix, combined with a machine learning model, which can effectively improve the recognition accuracy.

[0057] 3. Reduced computational complexity

[0058] Compared with cell cluster recognition based on image processing, the present invention directly models the pulse matrix without the need to store and process a large number of images, and has higher computational efficiency.

[0059] At the same time, compared with the method of multiple traversals and threshold judgment, the machine learning model can be calculated in parallel and the prediction speed is faster. The method of the present invention is efficient in operation.

[0060] 4. Improved robustness and ability to adapt to different samples

[0061] Machine learning models can obtain better generalization capabilities through pre-training and adapt to different parameter changes.

[0062] The model in the present invention is obtained based on pre-training of diverse samples, can identify cell clusters of different sizes and types, has strong robustness, and does not require parameter adjustment for each sample.

[0063] 5. Lay the foundation for standardized automated processing

[0064] The present invention realizes the standardization and automation of the step of cell cluster identification. Flow cytometry analysis can establish a standard end-to-end process and realize automation from collection to analysis results.

[0065] Users only need to perform simple operations to obtain accurate results after removing cell clusters, which greatly improves detection efficiency.

[0066] 6. Reduced testing costs

[0067] The improved automation can reduce the workload of professional operators. At the same time, the higher the calculation efficiency, the shorter the instrument usage time.

[0068] The application of the present invention can reduce the manpower and equipment costs of flow cytometry detection and improve detection efficiency.

[0069] In summary, the present invention can solve the technical problems that the existing cell cluster identification and removal methods rely on manually set thresholds and parameters, require manual adjustment of identification rules for different samples, cannot be automated, and the image-based methods have large storage and calculation requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 The present invention provides a flow chart of a method for outputting a cell removal signal by a flow cytometer. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described.

[0072] like Figure 1 FIG. 1 is a flowchart of a method for outputting a cell removal signal by a flow cytometer according to a first aspect of the present invention. The method comprises the following steps:

[0073] S10, collecting a set of electric pulse signals collected by the photoelectric converter of the flow cytometer, including a set of electric pulse signals generated by forward scattered light, a set of electric pulse signals generated by side scattered light, and a set of electric pulse signals generated by fluorescence signals, which are respectively recorded as FSC pulse set, SSC pulse set, and fluorescence pulse set; wherein the electric pulse signal includes time, length, width, and area;

[0074] S20, establishing an FSC matrix, an SSC pulse matrix, and a fluorescence pulse matrix respectively using the obtained FSC pulse set, SSC pulse set, and fluorescence pulse set according to the electrical pulse signal;

[0075] S30, traversing each electrical pulse signal in the FSC pulse set, acquiring electrical pulse signals whose channel values ​​are greater than a first threshold, and forming an FSC large pulse set;

[0076] S40, screening the FSC matrix, the SSC pulse matrix, and the fluorescence pulse matrix according to the FSC large pulse set to obtain a screening FSC matrix, a screening SSC pulse matrix, and a screening fluorescence pulse matrix;

[0077] S50, establishing a polymerization matrix using the obtained screening FSC matrix, screening SSC pulse matrix, and screening fluorescence pulse matrix;

[0078] S60, analyzing the aggregation matrix using a pre-trained cell cluster analysis model to obtain electrical pulse signals corresponding to the cell clusters in the aggregation matrix;

[0079] S70, deleting the electrical pulse signal corresponding to the cell cluster from the electrical pulse signal set, and outputting the obtained electrical pulse signal set to the data processing system of the flow cytometer.

[0080] Specifically, the implementation steps of S10 are described as follows:

[0081] In this step, we collect the original electrical signals collected by the photoelectric converter of the flow cytometer. The photoelectric converter generally uses devices such as photodiodes or phototransistors. When the cell sample flows through the laser beam, forward scattered light, side scattered light and fluorescence signals are generated. These light signals are converted into current signals by the photodiode, and then converted into voltage signals by the operational amplifier, which are the original electrical pulse signals.

[0082] The raw electrical pulse signal contains all the optical information about each cell or non-cell event. Different types of electrical pulse signals can be obtained according to the scattering and fluorescence characteristics of the cell or non-cell event in different channels. In the present invention, we divide these electrical pulse signals into three categories:

[0083] (1) FSC pulse set: The set of electrical pulse signals generated by forward scattered light (FSC). FSC mainly reflects the size of cells.

[0084] (2) SSC pulse set: The set of electrical pulse signals generated by side scattered light (SSC). SSC mainly reflects the internal structure and complexity of cells.

[0085] (3) Fluorescence pulse set: The set of electrical pulse signals generated by the fluorescence detected by different fluorescence channels. Different fluorescence channels correspond to different fluorescent markers.

[0086] The above electrical pulse signal contains the following information:

[0087] (1) Time: The time when the pulse signal appears.

[0088] (2) Length: The length of the pulse signal, i.e. the span in the X-axis direction. It reflects the amount of photon flux.

[0089] (3) Width: The width of the pulse signal, i.e. the height along the Y axis. It is related to the amplitude of the pulse signal.

[0090] (4) Area: The area under the pulse signal curve. Related to the number of photons detected.

[0091] In summary, the key to this step is to collect the original electrical pulse signals and classify them into FSC pulse sets, SSC pulse sets and fluorescence pulse sets according to different optical channels. These signals contain the optical characteristic information of each cell / non-cell event, laying the foundation for the subsequent identification of cell clusters.

[0092] Below we elaborate on the implementation process of this step:

[0093] 1. Configure the optical path system of the flow cytometer, including laser, focusing lens, flow chamber, etc., so that when the laser irradiates the sample, it can generate forward scattered light, side scattered light and fluorescence signals.

[0094] 2. Select appropriate photodiodes or phototransistors as photoelectric converters and configure an appropriate number according to actual needs to ensure that forward scattered light, side scattered light and signals of each fluorescence channel can be detected.

[0095] 3. Connect the output of the photodiode or phototransistor to an operational amplifier and configure the gain to amplify the current signal.

[0096] 4. Connect the digital acquisition card and configure the sampling frequency to collect the voltage signal output by the operational amplifier.

[0097] 5. Run the flow cytometer, using a cell sample with a predetermined fluorescent marker. When the cell sample flows through the laser beam, a light signal is generated.

[0098] 6. The photodiode converts the light signal into a current signal, and the operational amplifier amplifies the signal and converts it into a voltage pulse signal.

[0099] 7. The digital acquisition card samples the voltage pulse signal at a certain frequency to obtain the original electrical pulse signal.

[0100] 8. Based on the position of the photodiode, it can be determined whether the electrical pulse signal comes from FSC, SSC or a specific fluorescence channel.

[0101] 9. According to the above judgment, the collected original electrical pulse signals are classified into FSC pulse set, SSC pulse set and fluorescence pulse set.

[0102] 10. Analyze the electrical pulse signal and extract characteristic information such as time, length, width and area to provide a basis for subsequent identification of cell clusters.

[0103] 11. Store the processed electrical pulse signal in the computer memory or hard disk for subsequent use.

[0104] 12. Repeat steps 5-11 to collect a sufficient number of cell samples and obtain a set of original electrical pulse signals containing cell cluster information.

[0105] Through the above steps, we can effectively obtain the original electrical pulse signal collected by the photoelectric converter of the flow cytometer, classify it according to different optical channels, extract characteristic information, lay the foundation for the subsequent identification of cell clusters, and complete the implementation of step S10. Points to note include:

[0106] (1) The optical system should be reasonably configured to ensure effective detection of isotropic scattered light and fluorescence.

[0107] (2) The photoelectric converter must be selected correctly and the response speed must be fast enough.

[0108] (3) The magnification should be appropriate to prevent waveform distortion.

[0109] (4) The sampling frequency should be set reasonably to satisfy the Nyquist sampling theorem.

[0110] (5) The extracted features should be able to effectively represent the signal characteristics.

[0111] (6) Storage must ensure access speed.

[0112] By paying attention to the above points, we can obtain high-quality original electrical pulse signals containing rich information, provide reliable basic data for subsequent cell cluster identification, and complete the collection and processing of step S10.

[0113] Among them, in the above technical scheme, the step of establishing FSC matrix, SSC pulse matrix and fluorescence pulse matrix respectively according to the obtained FSC pulse set, SSC pulse set and fluorescence pulse set of the electrical pulse signal also includes the step of eliminating noise points of the electrical pulse signal.

[0114] Specifically, the implementation steps of S20 are described as follows:

[0115] First, define some variables and constants:

[0116] -FSC_pulse_set represents the FSC pulse set, which contains m pulse signals, denoted as {pSc1, psc2, ..., pSc m};

[0117] -SSC_pulse_set represents the SSC pulse set, which contains n pulse signals, denoted as {pssc1, pssc2, ..., pssc n};

[0118] -FL_pulse_set represents a fluorescence pulse set, which contains 1 pulse signal, denoted as {pfl1, pfl2, ..., pfl l};

[0119] -FSC_matrix indicates the FSC matrix to be established, with a dimension of m×4;

[0120] -SSC_matrix indicates the SSC matrix to be created, with a dimension of n×4;

[0121] -FL_matrix indicates the fluorescence matrix to be established, with a dimension of l×4;

[0122] - The pulse signal p is composed of four features: p = {t, w, a, h}, where t represents the pulse time, w represents the pulse width, a represents the pulse area, and h represents the pulse height;

[0123] Then, traverse each pulse signal psc in FSC_pulse_set i , write it into the i-th row of FSC_matrix;

[0124] Similarly, traverse SSC_pulse_set and FL_pulse_set and write the pulse signal into the corresponding matrix;

[0125] So far, the FSC matrix, SSC matrix and fluorescence matrix have been established according to the pulse signal set.

[0126] Preferably, in order to further eliminate the influence of noise points, the matrix can be smoothed:

[0127] 1) FSC matrix smoothing

[0128] Calculate the psc of each pulse i The distance d(psc i ):

[0129]

[0130] If d(psc i )<threshold, then psc i and psc i+1 Combined into one pulse.

[0131] Repeat the above steps until d(psc i )≥threshold.

[0132] 2) SSC matrix and fluorescence matrix smoothing are done in the same way as FSC matrix smoothing.

[0133] 3) Perform mean filtering on the smoothed matrix to further eliminate random noise.

[0134] FSC_matrix=smooth(FSC_matrix);

[0135] SSC_matrix=smooth(SSC_matrix);

[0136] FL_matrix=smooth(FL_matrix);

[0137] After the above steps, we obtain the smoothed FSC matrix, SSC matrix and fluorescence matrix. These matrices reflect the pulse distribution of each channel and lay the foundation for subsequent identification and analysis.

[0138] The above matrix processing algorithm uses smoothing and filtering methods, which can effectively eliminate random errors introduced by instrument noise and improve the accuracy of subsequent identification. Its time complexity is O(m+n+l) and its space complexity is O(m+n+l). When there are many cell samples, the matrix size will become larger. You can consider using methods such as PCA to reduce the dimension to reduce the amount of calculation. PCA (Principal Component Analysis) is a common data analysis method. This method uses orthogonal transformation to convert data represented by linearly related variables into data represented by a few linearly independent variables. Linearly independent variables are called principal components. The number of principal components is usually less than the number of original variables, so principal component analysis is often used to reduce the dimension of high-dimensional data and extract the main characteristic components of the data.

[0139] In addition, the pulse merging threshold needs to be determined based on the experience of specific instrument parameters, and is generally taken as 10-20% of the pulse width. When smoothing the matrix, mean filtering, median filtering or Gaussian filtering can be used, and the filter parameters need to be selected according to the actual situation.

[0140] In summary, this step traverses the pulse signal, establishes the pulse matrix, and performs smoothing filtering to lay the foundation for subsequent recognition and analysis. The algorithm is clear and direct, the amount of calculation is moderate, and a good pulse expression can be obtained. This method can be widely used in pulse signal processing in fields such as flow cytometry.

[0141] Among them, in the above technical solution, the step of traversing each electrical pulse signal in the FSC pulse set, obtaining the electrical pulse signal whose channel value is greater than the first threshold, and forming the FSC large pulse set specifically includes:

[0142] Traversing the FSC pulse set, extracting the pulse signal whose FSC channel value is greater than the first threshold as the first pulse set;

[0143] Calculating the inter-pulse distance of the first pulse set and merging adjacent pulses;

[0144] Performing smoothing filtering on the first pulse set after merging adjacent pulses;

[0145] The PCA method is used to reduce the dimension of the first pulse set after smoothing filtering.

[0146] The first pulse set after dimension reduction is used as the FSC large pulse set.

[0147] Specifically, the implementation steps of S30 are described as follows:

[0148] 1. Define variables and constants

[0149] -threshold1: FSC channel first threshold

[0150] -FSC_large_set: FSC large pulse set to be formed

[0151] 2. Traverse each pulse signal psc in FSC_pulse_set i , if psc i .h>threshold1, then psc i Join FSC_large_set;

[0152] 3. Calculate the time distance between each pulse signal in FSC_large_set. If the distance is less than threshold2, merge the two pulse signals.

[0153] 4. Smoothing Filter

[0154] Apply mean filtering and median filtering to smooth FSC_large_set.

[0155] 5. Dimensionality reduction

[0156] PCA and other methods can be used to reduce the dimension of FSC_large_set and reduce redundant features.

[0157] After the above steps, we extracted the large pulse signals from the FSC pulse set to form the FSC large pulse set FSC_large_set, in preparation for subsequent cell cluster identification and analysis.

[0158] By default, the first threshold, threshold1, is set to 3 times the FSC baseline noise;

[0159] Furthermore, the first threshold value threshold1 is used to extract the pulse signal with a larger FSC channel intensity from the FSC pulse set. Its determination needs to consider the following factors:

[0160] 1. Instrument parameters

[0161] The baseline noise and sensitivity of the FSC channels of different flow cytometers will vary, and an empirical range needs to be pre-determined based on the instrument parameters.

[0162] 2. Sample Type

[0163] Different types of cell samples have different size distributions, resulting in different FSC intensity distribution ranges, and the threshold needs to be adjusted according to the sample.

[0164] 3. Data Distribution

[0165] Visualize or count the data distribution of FSC pulse sets and determine a threshold that can effectively distinguish strong pulses from weak pulses.

[0166] 4. Trial and Error

[0167] You can select several candidate thresholds from the empirical range, test their screening effects with samples, and select the optimal threshold.

[0168] 5. Adaptive Threshold

[0169] A data-driven adaptive threshold algorithm can be designed, such as the Otsu method, which automatically determines the threshold based on the distribution of FSC pulse signals.

[0170] Taking all the above factors into consideration, it is generally recommended that the first threshold value, threshold1, be set within the range of 3-5 times the FSC baseline noise. The specific value needs to be determined for the sample and instrument. At the same time, the threshold value can also be optimized by trial and error, or an automatic threshold selection algorithm can be designed.

[0171] In step S30, threshold2 is used to determine whether the distance between two FSC large pulses is close enough to decide whether to merge them into one pulse. The default threshold2 is 3 times the pulse time length;

[0172] Of course, you can also determine threshold2 according to the following method. The following factors need to be considered:

[0173] 1. Instrument collection frequency

[0174] The acquisition frequency of the flow cytometer determines the minimum time interval between two consecutive pulse events. If threshold2 is lower than this interval, incorrect merging will occur.

[0175] 2. Physical width of cells or cell clusters

[0176] Consider the physical width of the cell cluster to avoid merging different cell clusters due to too small threshold2.

[0177] 3. Noise jitter effect

[0178] A threshold2 that is too small may also mistakenly combine noise jitter into a pulse.

[0179] 4. Statistical distribution of multi-sample distances

[0180] The distribution of pulse distances in multiple samples can be observed, and an appropriate interval can be selected as threshold2.

[0181] 5. Experience parameter adjustment

[0182] Manually test different parameters and select the threshold with the best recognition effect.

[0183] Taking all the above factors into consideration, the recommended setting of threshold2 is: 3-5 times of the acquisition frequency corresponding to the time; it can also be tested step by step from 0.5ms and above to select the optimal threshold that can effectively merge the cell cluster pulses without introducing over-merging.

[0184] Among them, in the above technical solution, the FSC matrix, the SSC pulse matrix and the fluorescence pulse matrix are screened according to the FSC large pulse set to obtain the steps of screening the FSC matrix, screening the SSC pulse matrix and screening the fluorescence pulse matrix, which specifically include:

[0185] Traverse the FSC matrix and keep the rows corresponding to the large FSC pulses;

[0186] According to the large FSC pulse, the corresponding row was screened from the SSC and fluorescence matrix;

[0187] The matrix after smoothing filter screening;

[0188] Output the smoothed FSC, SSC and fluorescence matrix.

[0189] Specifically, the implementation steps of S40 are described as follows:

[0190] 1. Define variables and constants

[0191] -FSC_large_set: FSC large pulse set obtained in step S30;

[0192] -FSC_matrix: FSC matrix obtained in step S20;

[0193] -SSC_matrix: SSC matrix obtained in step S20;

[0194] -FL_matrix: fluorescence matrix obtained in step S20;

[0195] -FSC_filter: FSC matrix after filtering;

[0196] -SSC_filter: SSC matrix after filtering;

[0197] -FL_filter: filtered fluorescence matrix;

[0198] 2. Screening FSC Matrix

[0199] Traverse the FSC_matrix row by row. If the pulse signal psc of the current row i If the row exists in FSC_large_set, keep the row; otherwise, delete the row.

[0200] 3. According to the same principle, screen SSC and fluorescence matrix

[0201] 4. Smoothing Filter

[0202] Perform smoothing filtering on the filtered matrix.

[0203] FSC_filter=smooth(FSC_filter);

[0204] SSC_filter=smooth(SSC_filter);

[0205] FL_filter=smooth(FL_filter);

[0206] After the above steps, we use the FSC large pulse set to screen other matrices and obtain the screened smooth matrix to prepare for the next step of identification.

[0207] Among them, in the above technical solution, the step of establishing a polymerization matrix using the obtained screening FSC matrix, the screening SSC pulse matrix and the screening fluorescence pulse matrix specifically includes:

[0208] The screening smoothed matrices of different channels were horizontally spliced ​​to form an aggregate matrix;

[0209] Normalize the aggregation matrix;

[0210] The normalized aggregation matrix is ​​subjected to dimensionality reduction using PCA to obtain a dimensionality-reduced aggregation matrix that integrates information from different channels.

[0211] Specifically, the implementation steps of S50 are described as follows:

[0212] 1. Define variables

[0213] -fusion_matrix: The aggregation matrix to be created

[0214] 2. Matrix aggregation

[0215] Concatenate the three matrices horizontally by row to get the fusion_matrix:

[0216] 3. Standardization

[0217] Normalize the columns of fusion_matrix:

[0218] Normalization can eliminate the effects of different feature magnitudes.

[0219] 4. Dimensionality reduction

[0220] You can use methods such as PCA to reduce the dimension of fusion_matrix:

[0221] At this point, we aggregate the feature matrices of different channels into an aggregate matrix, and perform standardization and dimensionality reduction processing to prepare for the next step of recognition.

[0222] Among them, in the above technical solution, the steps of establishing and training the cell cluster analysis model specifically include:

[0223] Collect multiple historical aggregation matrix samples and use them as training samples after labeling the cell cluster data;

[0224] Build machine learning classification models;

[0225] The training samples are used to train the machine learning classification model to obtain a cell cluster analysis model.

[0226] The specific description is as follows:

[0227] 1. Collect multiple historical aggregation matrix samples and mark them;

[0228] - Collect aggregated matrix samples containing single cells and cell clusters generated by multiple S50 steps;

[0229] - For each sample, manually determine the rows corresponding to single cells and cell clusters;

[0230] 2. Build a classification model

[0231] - Input each matrix row as a sample;

[0232] -Use the marked categories as labels;

[0233] -Build machine learning classification models, such as SVM, random forest, etc.;

[0234] 3. Training the model

[0235] - Divide the training set and validation set;

[0236] -Train classification models and optimize model hyperparameters;

[0237] -Evaluate the model performance on the validation set;

[0238] 4. Further, it also includes model evaluation

[0239] - Evaluate classification performance on an independent test set;

[0240] -Calculate indicators such as precision and recall.

[0241] Furthermore, in the above technical solution, the model used to construct the machine learning classification model is SVM or random forest.

[0242] Among them, in the above technical solution, the step of deleting the electric pulse signal corresponding to the cell cluster in the electric pulse signal set specifically includes:

[0243] Traversing the original electrical pulse signal set;

[0244] Determine whether each pulse is in the identified cell cluster set;

[0245] If so, delete the pulse from the original set;

[0246] The pulses of non-cell clusters are retained to form a filtered output set;

[0247] Returns the filtered electrical pulse signal set.

[0248] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for outputting a decellularization signal by a flow cytometer.

[0249] A third aspect of the present invention provides a system for outputting a decellularization signal by a flow cytometer, which includes the above-mentioned computer-readable storage medium.

[0250] Specifically, the principle of the present invention is:

[0251] 1. Multi-channel information fusion

[0252] Flow cytometer can simultaneously obtain multiple parameters of each cell or cell cluster, such as FSC, SSC, fluorescence, etc. The present invention fuses the information matrices of different channels to form a comprehensive multi-feature aggregation matrix. Since there is a certain connection between FSC, SSC, and fluorescence signals of each cell or cell cluster, such as the cell cluster is relatively large in volume and has a more complex internal structure, FSC and SSC will have a certain correlation. At the same time, the fluorescence signal of the cell cluster will also produce a different channel value from that of ordinary cells due to the relatively large number of fluorescent dyes in the cell cluster. If manual judgment is simply adopted, the selected threshold value will not be accurate enough due to the large amount of data. Therefore, the method of multi-channel fusion is used to perform multi-feature extraction and analysis using the aggregation matrix, so that the cell cluster signal can be better determined.

[0253] Compared with a single parameter, the aggregation matrix provides richer features to describe each event, which is beneficial for subsequent recognition and classification. The fusion of multimodal information is an important way to improve the expressive ability.

[0254] 2. Matrix representation encoding slight changes

[0255] The present invention adopts a matrix form to organize the description indicators of pulse events, such as pulse shape, time and other information.

[0256] Compared with simple statistics, the matrix preserves the slight changes of the original pulses, especially the relationship between signals in different channels, providing richer features for cell cluster identification.

[0257] 3. Pre-trained machine learning models

[0258] The present invention uses a pre-trained machine learning model to classify and analyze the aggregation matrix to identify the cell clusters therein.

[0259] Compared with fixed thresholds and manual judgment rules, machine learning models can learn the complex distribution of samples and have stronger generalization capabilities for unknown data.

[0260] 4. End-to-end automated process

[0261] From the original signal to the final output, the present invention establishes a completely automated analysis process without the need for manual intervention and parameter setting.

[0262] Standardized end-to-end processes are the key to achieving automated testing.

[0263] 5. Balance accuracy and efficiency

[0264] The present invention takes into account both recognition accuracy and computational efficiency: first, multi-channel aggregation is performed to increase the amount of information, and then algorithmic dimensionality reduction is used to ensure efficiency.

[0265] The balance between accuracy and efficiency is also an important indicator of the automated detection system.

[0266] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for outputting a cell removal signal by a flow cytometer, characterized in that: The following steps are involved: S10, collecting a set of electric pulse signals collected by the photoelectric converter of the flow cytometer, including a set of electric pulse signals generated by forward scattered light, a set of electric pulse signals generated by side scattered light, and a set of electric pulse signals generated by fluorescence signals, which are respectively recorded as FSC pulse set, SSC pulse set, and fluorescence pulse set; wherein the electric pulse signal includes time, length, width, and area; S20, establishing an FSC matrix, an SSC pulse matrix, and a fluorescence pulse matrix respectively using the obtained FSC pulse set, SSC pulse set, and fluorescence pulse set according to the electrical pulse signal; S30, traversing each electrical pulse signal in the FSC pulse set, acquiring electrical pulse signals whose channel values ​​are greater than a first threshold, and forming an FSC large pulse set; S40, screening the FSC matrix, the SSC pulse matrix, and the fluorescence pulse matrix according to the FSC large pulse set to obtain a screening FSC matrix, a screening SSC pulse matrix, and a screening fluorescence pulse matrix; S50, establishing a polymerization matrix using the obtained screening FSC matrix, screening SSC pulse matrix, and screening fluorescence pulse matrix; S60, analyzing the aggregation matrix using a pre-trained cell cluster analysis model to obtain electrical pulse signals corresponding to the cell clusters in the aggregation matrix; S70, deleting the electrical pulse signal corresponding to the cell cluster from the electrical pulse signal set, and outputting the obtained electrical pulse signal set to the data processing system of the flow cytometer.

2. The method for outputting a cell removal signal by a flow cytometer according to claim 1, characterized in that: The step of establishing the FSC matrix, SSC pulse matrix and fluorescence pulse matrix respectively using the obtained FSC pulse set, SSC pulse set and fluorescence pulse set according to the electrical pulse signal also includes the step of eliminating noise points of the electrical pulse signal.

3. The method for outputting a cell removal signal by a flow cytometer according to claim 1, characterized in that: The step of traversing each electrical pulse signal in the FSC pulse set, acquiring the electrical pulse signal whose channel value is greater than the first threshold, and forming the FSC large pulse set specifically includes: Traversing the FSC pulse set, extracting the pulse signal whose FSC channel value is greater than the first threshold as the first pulse set; Calculate the inter-pulse distance of the first pulse set and merge adjacent pulses; Performing smoothing filtering on the first pulse set after merging adjacent pulses; The PCA method is used to reduce the dimension of the first pulse set after smoothing filtering. The first pulse set after dimension reduction is used as the FSC large pulse set.

4. The method for outputting a cell removal signal by a flow cytometer according to claim 1, characterized in that: The steps of screening the FSC matrix, the SSC pulse matrix and the fluorescence pulse matrix according to the FSC large pulse set to obtain the steps of screening the FSC matrix, screening the SSC pulse matrix and screening the fluorescence pulse matrix specifically include: Traverse the FSC matrix and keep the rows corresponding to the large FSC pulses; According to the large FSC pulse, the corresponding row was screened from the SSC and fluorescence matrix; The matrix after smoothing filter screening; Output the smoothed FSC, SSC and fluorescence matrix.

5. The method for outputting a cell removal signal by a flow cytometer according to claim 1, characterized in that: The step of establishing a polymerization matrix using the obtained screening FSC matrix, screening SSC pulse matrix and screening fluorescence pulse matrix specifically includes: The screening smoothed matrices of different channels were horizontally spliced ​​to form an aggregate matrix; Normalize the aggregation matrix; The normalized aggregation matrix is ​​subjected to dimensionality reduction using PCA to obtain a dimensionality-reduced aggregation matrix that integrates information from different channels.

6. The method for outputting a cell removal signal by a flow cytometer according to claim 1, characterized in that: The steps of establishing and training the cell cluster analysis model specifically include: Collect multiple historical aggregation matrix samples and use them as training samples after labeling the cell cluster data; Build machine learning classification models; The training samples are used to train the machine learning classification model to obtain a cell cluster analysis model.

7. The method for outputting a cell removal signal by a flow cytometer according to claim 6, characterized in that: The model used to build the machine learning classification model is SVM or random forest.

8. The method for outputting a cell removal signal by a flow cytometer according to claim 1, characterized in that: The step of deleting the electric pulse signal corresponding to the cell cluster from the electric pulse signal set specifically includes: Traversing the original electrical pulse signal set; Determine whether each pulse is in the identified cell cluster set; If so, delete the pulse from the original set; The pulses of non-cell clusters are retained to form a filtered output set; Returns the filtered electrical pulse signal set.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for outputting a decellularization signal by a flow cytometer according to any one of claims 1 to 8.

10. A system for outputting a cell removal signal by a flow cytometer, characterized in that: Includes the computer-readable storage medium of claim 9.

Images