A method, medium and system for analyzing flow cytometry detection data

Abstract

The present invention provides a method, medium and system for analyzing flow cytometry detection data, belonging to the technical field of flow cytometry. The method includes: collecting an electric pulse signal set collected by a photoelectric converter of a flow cytometer; performing preprocessing on the electric pulse signal set, including denoising, filtering and enhancement; calibrating abnormal signals in the electric pulse signal set, and splitting the electric pulse signal set into a normal signal set and an abnormal signal set; using a pre-trained normal signal analysis model to analyze the normal signal set to obtain a representative cell set of a detection sample; performing classification analysis on the abnormal signal set to obtain a representative cell adjustment factor, and adjusting the quantity ratio of each representative cell in the representative cell set. The present invention can solve the technical problem that the obtained flow cytometry detection data will have batch effects, the statistical feature distributions of cells in different batches are different, and the shape of sample cells cannot be described more accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of flow cytometers, and specifically relates to a method, medium, and system for analyzing flow cytometer detection data. Background Art

[0002] Flow cytometers are widely used in the fields of clinical diagnosis, scientific research, drug development, etc. A flow cytometer is an automated instrument that can quickly perform quantitative and qualitative analysis on cell samples. It is based on the principle of laser scattering. When a single cell passes through the laser irradiation, forward scattering, side scattering, and fluorescence signals will occur. These optical signals are converted into electrical pulse signals by a photoelectric sensor for acquisition. Through statistical analysis, statistical characteristics such as cell size, internal complexity, and fluorescence intensity in the sample can be obtained.

[0003] However, due to the influence of fluid force on cells in the flow cytometer channel, there are different flow postures instead of a fixed shape, and the postures of different cells in the channel are different, which will affect the detection of optical signals. For example, some cells are relatively flat, and different angles or positions of laser irradiation on the cells will cause differences in the generated forward scattering, side scattering, and fluorescence signals. Therefore, there will be batch effects in the obtained flow cytometry detection data, and the statistical characteristic distributions of cells in different batches are different, and the shape of the sample cells cannot be described more accurately. Summary of the Invention

[0004] In view of this, the present invention provides a method, medium, and system for analyzing flow cytometer detection data, which can solve the technical problems that there are batch effects in the obtained flow cytometry detection data, the statistical characteristic distributions of cells in different batches are different, and the shape of the sample cells cannot be described more accurately.

[0005] The present invention is implemented as follows:

[0006] The first aspect of the present invention provides a method for analyzing flow cytometer detection data, which includes the following steps:

[0007] S10. Collect the set of electrical pulse signals collected by the photoelectric converter of the flow cytometer, including the set of electrical pulse signals generated by forward scattered light, the set of electrical pulse signals generated by side scattered light, and the set of electrical pulse signals generated by fluorescence signals, which are respectively denoted as the FSC pulse set, the SSC pulse set, and the fluorescence pulse set;

[0008] S20. Perform preprocessing on the set of electrical pulse signals, including denoising, filtering, and enhancement;

[0009] S30. Calibrate the abnormal signals in the set of electrical pulse signals, and split the set of electrical pulse signals into a set of normal signals and a set of abnormal signals;

[0010] S40. Analyze the set of conventional signals using a pre-trained conventional signal analysis model to obtain a representative cell set of the test sample. The representative cell set includes 10 to 20 cells of the same type as the test sample and the quantity ratio of each representative cell in the representative cell set;

[0011] S50. Conduct a classification analysis on the set of abnormal signals to obtain a representative cell adjustment factor, and use the representative cell adjustment factor to adjust the quantity ratio of each representative cell in the representative cell set;

[0012] S60. Output the obtained representative cell set and the quantity ratio of each representative cell therein to the experimenter.

[0013] Based on the above technical solution, a method for analyzing flow cytometer detection data of the present invention can also be improved as follows:

[0014] Among them, the denoising method is median filtering, the filtering method is band-pass filtering, and the enhancement method is contrast-limited adaptive histogram equalization.

[0015] Among them, the step of calibrating abnormal signals in the set of electrical pulse signals and splitting the set of electrical pulse signals into a set of conventional signals and a set of abnormal signals specifically includes:

[0016] Detect bright abnormal signals or noise abnormal signals in the set of electrical pulse signals;

[0017] Calibrate the detected bright abnormal signals or noise abnormal signals, and record the calibration information;

[0018] Split the signal set into a conventional set and an abnormal set according to the calibration information.

[0019] Among them, the steps of establishing and training the conventional signal analysis model specifically include:

[0020] Establish a training data set and a fine-tuning data set. The training data set is the set of electrical pulse signals collected by using eight-peak rainbow balls for flow cytometer detection as the training electrical pulse signal set. The data obtained by preprocessing the training electrical pulse signal set including denoising, filtering, and enhancement is used as the input for training, and the output of training is the size of the eight-peak rainbow balls;

[0021] The fine-tuning dataset is a set of electrical pulse signals obtained by performing flow cytometry on a small sample of the same type of cells in the sample detected by the flow cytometer. The data obtained after preprocessing the fine-tuning electrical pulse signal set, including denoising, filtering, and enhancement, is used as the input for fine-tuning. The output of fine-tuning is a representative cell set of the small sample. The representative cell set includes 10 to 20 cells within the small sample. The representative cell set also includes the proportion of each representative cell. The number of cells in the small sample is M times that of 106, where M ranges from 5 to 10;

[0022] Use a convolutional neural network to establish a prototype of a conventional signal analysis model, and use the training dataset to train the prototype of the conventional signal analysis model to obtain a conventional signal analysis model;

[0023] Fine-tune the conventional signal analysis model using the fine-tuning data, and use the fine-tuned conventional signal analysis model as the output conventional signal analysis model.

[0024] Furthermore, the steps to obtain the representative cell set of the small sample and the proportion of each representative cell are as follows:

[0025] Step 1: Take a small amount of cell samples to be detected, and set the number of cells to K; K ranges from 5 to 10 times the total number of cells to be detected;

[0026] Step 2: Directly observe the taken samples using a microscope and manually count the number of cells with different morphologies. Let the number of different cell categories observed be W, where W ranges from 10 to 20;

[0027] Step 3: For each cell category, count its proportion in the sample. Let the number of the i-th type of cells be w i , then the proportion of the i-th type of cells is p i =w i / K;

[0028] Step 4: According to the morphological characteristics of the cells, select M types of representative cells to form a representative cell set R = r1, r2, …, r W ;

[0029] Step 5: Construct the corresponding quantity proportion P = p1, p2, …, p W ;

[0030] Step 6: Take the representative cell set R and its proportion P obtained from the observation and statistics as the representative cell set and proportion of the small sample cells.

[0031] Among them, the method for classifying and analyzing the abnormal signal set is the decision tree algorithm.

[0032] Further, the electric pulse signal includes time, length, width, and area.

[0033] Further, the way to fine-tune the conventional signal analysis model using the fine-tuning data is to only fine-tune the last three layers of the conventional signal analysis model.

[0034] The second aspect of the present invention provides a computer-readable storage medium, wherein program instructions are stored in the computer-readable storage medium, and when the program instructions run, they are used to execute the above-mentioned data analysis method for flow cytometry detection.

[0035] The third aspect of the present invention provides a data analysis system for flow cytometry detection, which includes the above-mentioned computer-readable storage medium.

[0036] Compared with the prior art, the beneficial effects of the data analysis method, medium, and system for flow cytometry detection provided by the present invention are as follows:

[0037] 1. By establishing a representative cell set, the influence of morphological changes of cells in different batches in the flow channel can be eliminated, making the recognition of cells more accurate and reliable.

[0038] 2. Using the representative cell set to describe a cell population can eliminate the influence of changes in the state of individual cells and more accurately depict the overall characteristics of the population.

[0039] 3. By means of deep learning methods for model training, the internal characteristics of cells with different morphologies can be learned, rather than relying solely on the statistical parameters obtained from flow cytometry detection, improving the recognition accuracy.

[0040] 4. By detecting and analyzing abnormal signals, the interference of noise such as cell debris on the detection results can be effectively filtered.

[0041] 5. Establishing a model training framework including fine-tuning can continuously optimize the model, adapt to the subtle changes between different experimental batches, and reduce batch effects.

[0042] 6. The proportion of each type of representative cell in the sample can be directly output for cell quantitative analysis.

[0043] In summary, the technical solution of the present invention solves the technical problems that the flow cytometry detection data obtained has batch effects, the statistical feature distributions of cells in different batches are different, and the shape of sample cells cannot be described more accurately. Description of the Drawings

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments of the present invention. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0045] Figure 1 It is a flowchart of a method for analyzing flow cytometer detection data provided by the present invention. Specific embodiments

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention.

[0047] As Figure 1 shown, it is a flowchart of a method for analyzing flow cytometer detection data provided by the first aspect of the present invention. The method includes the following steps:

[0048] S10. Collect the set of electrical pulse signals collected by the photoelectric converter of the flow cytometer, including the set of electrical pulse signals generated by forward scatter light, the set of electrical pulse signals generated by side scatter light, and the set of electrical pulse signals generated by fluorescence signals, which are respectively denoted as the FSC pulse set, the SSC pulse set, and the fluorescence pulse set;

[0049] S20. Perform preprocessing on the set of electrical pulse signals, including denoising, filtering, and enhancement;

[0050] S30. Calibrate the abnormal signals in the set of electrical pulse signals, and split the set of electrical pulse signals into a set of normal signals and a set of abnormal signals;

[0051] S40. Analyze the set of normal signals using a pre-trained normal signal analysis model to obtain a representative cell set of the test sample. The representative cell set includes 10 to 20 cells of the same type as the test sample and the quantity ratio of each representative cell in the representative cell set;

[0052] S50. Perform classification analysis on the set of abnormal signals to obtain a representative cell adjustment factor, and use the representative cell adjustment factor to adjust the quantity ratio of each representative cell in the representative cell set;

[0053] S60. Output the obtained representative cell set and the quantity ratio of each representative cell therein to the experimenter.

[0054] The specific implementation manner of step S10 is described as follows:

[0055] 1. The acquisition device is the photoelectric converter of a flow cytometer, which is used to acquire the electrical pulse signals generated by each cell flowing through the converter.

[0056] 2. Among them, the photoelectric converter of the flow cytometer at least includes a sensor for acquiring forward scatter light, a sensor for acquiring side scatter light, and a sensor for acquiring fluorescence.

[0057] 3. The electrical pulse signals acquired by each sensor respectively constitute an FSC pulse set, an SSC pulse set, and a fluorescence pulse set.

[0058] 4. Each electrical pulse signal contains four characteristic parameters: time, pulse width, pulse height (corresponding to signal intensity), and pulse area.

[0059] 5. For each passing cell, each of the three sensors can acquire an electrical pulse signal, and these three electrical pulse signals form the characteristic description of the cell.

[0060] 6. Repeat the above steps to acquire the electrical pulse signals of all flowing cells, and obtain a complete set of electrical pulse signals, including three parts: FSC, SSC, and fluorescence.

[0061] Among them, in the above technical solution, the denoising method is median filtering, the filtering method is band-pass filtering, and the enhancement method is contrast-limited adaptive histogram equalization.

[0062] Specifically, the implementation manner of step S20 is described as follows:

[0063] 1. For each electrical pulse signal in the input set of electrical pulse signals, perform denoising processing to eliminate the random error caused by background noise. It can be implemented by using the median filtering algorithm.

[0064] 2. Perform band-pass filtering on the denoised set of electrical pulse signals to filter out low-frequency random noise and high-frequency spike noise. Design appropriate filtering parameters to retain the effective electrical pulse signals.

[0065] 3. Perform contrast-limited adaptive histogram equalization (CLAHE) processing on the filtered set of electrical pulse signals to enhance the contrast of the electrical pulse signals and improve the effect of subsequent analysis.

[0066] 4. Perform smoothing processing on the enhanced set of electrical pulse signals. For example, use the Savitzky-Golay filter, which can effectively reduce noise.

[0067] 5. Optionally, a signal quality scoring method can be set to perform quality scoring on each processed electrical pulse signal, and eliminate the signals with lower quality scores.

[0068] 6. Output the preprocessed set of electrical pulse signals as the input for subsequent abnormal signal detection and signal analysis.

[0069] 7. Repeat the above steps to preprocess all input sets of electrical pulse signals and output the preprocessed results.

[0070] The above constitutes the specific implementation process of step S20, realizing the preprocessing of the input set of electrical pulse signals, including denoising, filtering, enhancement, etc., improving the accuracy of subsequent analysis.

[0071] Among them, in the above technical solution, the steps of calibrating the abnormal signals in the set of electrical pulse signals and splitting the set of electrical pulse signals into a set of normal signals and a set of abnormal signals specifically include:

[0072] Detect the brightness abnormal signals or noise abnormal signals in the set of electrical pulse signals;

[0073] Calibrate the detected brightness abnormal signals or noise abnormal signals and record the calibration information;

[0074] Split the signal set into a normal set and an abnormal set according to the calibration information.

[0075] Specifically, the implementation manner of step S30 is described as follows:

[0076] 1. Abnormal signal detection

[0077] To distinguish normal signals and abnormal signals, it is first necessary to detect the abnormal signals in the input set of electrical pulse signals. Common abnormal signals mainly include two aspects:

[0078] (1) Brightness abnormal signals

[0079] The brightness of the electrical pulse signal mainly depends on the brightness of the fluorescent dye and the laser intensity of the laser. If the fluorescent dye is not marked sufficiently, or the laser aging causes insufficient laser intensity, it will lead to a relatively weak overall brightness of the electrical pulse signal. On the other hand, if the filter is aging and the filtering effect is poor, or the gain of the electronic system is too high, it will lead to a relatively strong overall brightness of the electrical pulse signal.

[0080] Therefore, by analyzing the characteristic parameters of each electrical pulse signal, it is possible to judge whether its brightness is normal. Assume that the pulse peak voltage of the electrical pulse signal is V, and the empirical threshold range is [V min , V max , then the determination condition for detecting the brightness abnormal signal is:

[0081] V < V min or V > V max ;

[0082] (2) Abnormal noise signal

[0083] In a flow cytometer system, various noises are inevitably present, such as photoelectric noise, background noise, etc. These noises will cause large random errors in the electrical pulse signal. The abnormal noise can be detected by analyzing the difference between the peak value and the average value of the electrical pulse signal:

[0084] Let the values of n sampling points of the pulse signal be x1, x2, …, x n , then

[0085]

[0086] V peak =max(x1, x2, …, x n );

[0087] Then the detection condition for the abnormal noise signal is:

[0088] V peak -μ>σ th ;

[0089] Among them, σ th is a preset threshold.

[0090] 2. Abnormal signal calibration

[0091] Calibrate the detected abnormal signal so as to remove it from the set of electrical pulse signals. The calibration method can be to assign a specific flag value to the abnormal signal or directly record its subscript.

[0092] 3. Signal set segmentation

[0093] According to the result of the abnormal signal calibration, the set of electrical pulse signals is divided into two parts:

[0094] (1) Conventional signal set S normal

[0095] (2) Abnormal signal set S abnormal

[0096] The signals in both sets carry corresponding calibration information.

[0097] In this way, the detection and calibration of abnormal signals are realized, and the conventional signal set and the abnormal signal set are obtained by segmentation, which prepares for subsequent cell feature analysis.

[0098] Among them, in the above technical solution, the steps of establishing and training the conventional signal analysis model specifically include:

[0099] Establish a training dataset and a fine-tuning dataset. The training dataset is the set of electrical pulse signals collected by using eight-peak rainbow beads for flow cytometry detection. The set of electrical pulse signals collected is used as the training electrical pulse signal set. After preprocessing the training electrical pulse signal set, including denoising, filtering, and enhancement, the data is used as the training input, and the output of the training is the size of the eight-peak rainbow beads.

[0100] The fine-tuning dataset is the set of electrical pulse signals obtained by performing flow cytometry on a small sample of the same type of cells in the sample detected by the flow cytometry. This set of electrical pulse signals is used as the fine-tuning electrical pulse signal set. After preprocessing the fine-tuning electrical pulse signal set, including denoising, filtering, and enhancement, the data is used as the fine-tuning input, and the output of the fine-tuning is the representative cell set of the small sample. The representative cell set includes 10 - 20 cells within the small sample. The representative cell set also includes the proportion of each representative cell. The number of cells in the small sample is M times 106, where M ranges from 5 to 10.

[0101] Use a convolutional neural network to establish a prototype of a conventional signal analysis model, and use the training dataset to train the prototype of the conventional signal analysis model to obtain a conventional signal analysis model.

[0102] Fine-tune the conventional signal analysis model using the fine-tuning data, and use the fine-tuned conventional signal analysis model as the output conventional signal analysis model.

[0103] Specifically, the implementation method of the steps for establishing and training the conventional signal analysis model is described as follows:

[0104] 1. Establish a training dataset and a fine-tuning dataset

[0105] (1) Training dataset

[0106] Use standard eight-peak rainbow microspheres for flow cytometry detection to obtain a training electrical pulse signal set S containing multiple microsphere electrical pulse signals train . This dataset satisfies:

[0107] - Number of microspheres: N ball , take a sufficiently large value to ensure sufficient data volume

[0108] - True radius of each microsphere: r i , i = 1, 2, …, N ball ;

[0109] Perform preprocessing on S train to obtain a training input feature dataset X train . Correspondingly, the true radius of each microsphere constitutes a training label dataset Y train .

[0110] (2) Fine-tuning dataset

[0111] Select a small number (the number of cells is N cell ) of the same type of cells from the samples actually detected by the flow cytometer, and obtain the corresponding set S of fine-tuning electrical pulse signals fine . Perform preprocessing on it to obtain the fine-tuning input feature dataset X fine .

[0112] At the same time, statistically analyze the representative cell set R of these cells through microscopic observation fine ={r1, r2, …, r Z} and its quantity ratio P fine ={p1, p2, …, p Z}. Among them, the number Z of representative cells is taken as 10 - 20, and the total number N of cells cell is taken as 5 - 10 times the total number of cells to be detected

[0113] Then (X fine , R fine , P fine ) constitutes the fine-tuning dataset

[0114] 2. Establish the model prototype

[0115] Construct the prototype of the convolutional neural network model, including the convolutional layer, pooling layer, etc. The network structure is as follows

[0116] - Input layer: The input size is n×n

[0117] - Convolutional layer 1: The convolutional kernel size is k1, the number is c1, and the activation function is ReLU

[0118] - Pooling layer 1: The pooling size is p1

[0119] - Convolutional layer 2: The convolutional kernel size is k2, the number is c2, and the activation function is ReLU

[0120] - Pooling layer 2: The pooling size is p2

[0121] - Fully connected layer 1: The number of nodes is n1, and the activation function is ReLU

[0122] - Output layer: The number of nodes is 1, and the activation function is Linear

[0123] 3. Train the model

[0124] Use the training set (X train , Y train ) to train the network, and adopt the MSE loss function and Adam optimization algorithm. During the training process, save the model with the minimum loss on the validation set as the prototype of the conventional signal analysis model

[0125] 4. Fine-tune the model

[0126] Use fine-tuning set (X fine ,R fine ,P fine )Continue training to obtain a fine-tuned conventional signal analysis model.

[0127] Preferably, in order to increase the fine-tuning speed, you can load the model prototype, freeze the parameters of the previous layers, and only fine-tune the last 3 layers of the network.

[0128] 5. Output model

[0129] Finally, a pre-trained and fine-tuned conventional signal analysis model was obtained, which can analyze the input conventional electrical pulse signals and output a representative set of cells and the corresponding quantity ratio.

[0130] In summary, through the process of establishing a data set, building a model prototype, pre-training, and fine-tuning, a conventional signal analysis model was obtained. This model can effectively analyze conventional electrical pulse signals and extract representative features of cells.

[0131] Furthermore, in the above technical solution, the steps of obtaining a representative cell set of a small sample and the proportion of each representative cell thereof are as follows:

[0132] Step 1: Take a small amount of cell sample to be tested, and set the number of cells as K; the value of K is 5 to 10 times the total number of cells to be tested;

[0133] Step 2: Use a microscope to directly observe the sample, and manually count the number of cells with different morphologies observed. Let the number of different cell types observed be W, where W is 10 to 20;

[0134] Step 3: For each cell type, count its proportion in the sample, assuming the number of cells in the i-th type is w i , then the proportion of cells of type i is p i =w i / K;

[0135] Step 4: According to the morphological characteristics of the cells, select M representative cells to form a representative cell set R = r1, r2, ..., r W ;

[0136] Step 5: Construct the corresponding quantity ratio P = p1, p2, ..., p of each type of representative cells W ;

[0137] Step 6: The representative cell set R and its proportion P obtained through observation and statistics are used as the representative cell set and proportion of the small sample cells.

[0138] Among them, in the above technical solution, the method for classifying and analyzing the abnormal signal set is a decision tree algorithm.

[0139] Specifically, the specific implementation of step S50 is as follows:

[0140] 1. Input data

[0141] Use the abnormal signal set output by step S30 as the input, denoted as S abnormal .

[0142] 2. Load the model

[0143] Load the pre-trained abnormal signal classification model based on decision tree. This model can classify the input signal and output the signal category and proportion.

[0144] 3. Model prediction

[0145] Use the model to predict the abnormal signal set S abnormal to obtain the number N i and proportion p i of each type of abnormal signal.

[0146] 4. Calculate the adjustment factor

[0147] Assume that the number of detected abnormal signal categories is L, and the total number of all abnormal signals is N a , then the adjustment factor is calculated as:

[0148]

[0149] 5. Adjust the representative cell proportion

[0150] Load the representative cell set R = r1, r2,..., r M output by step S40, and the proportion P = p1, p2,..., p M .

[0151] The adjusted proportion P' = p'1, p'2,..., p' M satisfies:

[0152] p' i = p i × F;

[0153] 6. Output the result

[0154] Use the adjusted representative cell set R and its proportion P' as the output of step S50.

[0155] Through the above steps, the pre-trained decision tree model is used to analyze the abnormal signals, calculate the adjustment factor, and adjust the representative cell set proportionally, completing the implementation process of S50.

[0156] The specific implementation of step S60 is as follows:

[0157] 1. Aggregate the input results

[0158] Collect the representative cell set R and its quantity ratio P output by step S40, as well as the adjusted representative cell set R' and ratio P' output by step S50.

[0159] 2. Generate a report

[0160] Generate an analysis report based on R' and P'. The report includes:

[0161] (1) Detected representative cell categories

[0162] (2) The quantity and proportion of each type of representative cell

[0163] (3) The characteristic morphology of each type of representative cell shown in the image

[0164] 3. Output the report

[0165] Output the generated cell analysis report to the experimenter in a readable format (such as PDF).

[0166] 4. Interactive optimization

[0167] Obtain the experimenter's feedback on the report, and further optimize the selection criteria of representative cells according to the feedback.

[0168] 5. Save the results

[0169] Save the results such as the representative cell set, ratio, report, etc. into the database as the sample analysis record.

[0170] In summary, step S60 generates a report containing the details of representative cell analysis by integrating the results of the previous steps, outputs it to the user, and completes the process of flow cytometer signal analysis and cell quantitative analysis.

[0171] Furthermore, in the above technical solution, the electrical pulse signal includes time, length, width, and area.

[0172] Furthermore, in the above technical solution, the way of fine-tuning the conventional signal analysis model using the fine-tuning data is to only fine-tune the last three layers of the conventional signal analysis model.

[0173] Specifically, the principle of the present invention is:

[0174] 1. Establish a cell morphological feature model

[0175] The present invention constructs a representative cell set, and uses methods including deep learning to establish morphological feature models of various representative cells, which can more accurately describe the cell shape and overcome the influence of single-cell state changes.

[0176] 2. Abnormal signal detection and analysis

[0177] By detecting abnormal signals during the flow cytometry process, such as abnormal brightness, noise, etc., and performing analysis and processing, the error sources can be filtered, and the result reliability can be improved.

[0178] 3. Sample fine-tuning

[0179] Introducing small samples for model fine-tuning can make the model adapt to the subtle changes between different batches of samples and reduce the influence of batch effects.

[0180] 4. Multi-parameter detection constraint morphology discrimination

[0181] When a flow cytometer performs multi-parameter detection and combines the constraint conditions of these parameters, the cell morphology can be judged more accurately.

[0182] 5. Statistical evaluation of cell populations

[0183] By statistically evaluating the representative cell populations, the influence of the state changes of individual cells can be eliminated, and the characteristics of the cell population can be described more accurately.

[0184] 6. Direct output of cell quantitative results

[0185] Give the proportion of each type of representative cell for quantitative analysis, avoiding complex calculations.

[0186] In summary, through technical means such as modeling cell morphology, detecting abnormal signals, fine-tuning samples, multi-parameter constraints, and statistical evaluation of cell populations, the present invention can improve the accuracy of describing cell morphology, effectively reduce the error between different batches, and solve the batch effect problem existing in flow cytometry detection.

[0187] The above is only the specific implementation manner 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, and all of them should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for analyzing data detected by a flow cytometer, characterized in that, It includes the following steps: S10. Collect the set of electrical pulse signals collected by the photoelectric converter of the flow cytometer, including the set of electrical pulse signals generated by forward scatter light, the set of electrical pulse signals generated by side scatter light, and the set of electrical pulse signals generated by fluorescence signals, which are respectively denoted as the FSC pulse set, the SSC pulse set, and the fluorescence pulse set; S20. Perform preprocessing on the set of electrical pulse signals, including denoising, filtering, and enhancement; S30. Calibrate the abnormal signals in the set of electrical pulse signals, and split the set of electrical pulse signals into a normal signal set and an abnormal signal set; S40. Use the pre-trained normal signal analysis model to analyze the normal signal set to obtain the representative cell set of the test sample. The representative cell set includes 10 to 20 cells of the same type as the test sample and the quantity ratio of each representative cell in the representative cell set; S50. Perform classification analysis on the abnormal signal set to obtain the representative cell adjustment factor, and use the representative cell adjustment factor to adjust the quantity ratio of each representative cell in the representative cell set; S60. Output the obtained representative cell set and the quantity ratio of each representative cell therein to the experimenter.

2. The flow cytometer detection data analysis method according to claim 1, characterized in that The denoising method is median filtering, the filtering method is band-pass filtering, and the enhancement method is contrast-limited adaptive histogram equalization.

3. A method for analyzing data detected by a flow cytometer according to claim 1, characterized in that, The step of calibrating the abnormal signals in the set of electrical pulse signals and splitting the set of electrical pulse signals into a normal signal set and an abnormal signal set specifically includes: Detect the brightness abnormal signals or noise abnormal signals in the set of electrical pulse signals; Calibrate the detected brightness abnormal signals or noise abnormal signals, and record the calibration information; Divide the signal set into a normal set and an abnormal set according to the calibration information.

4. The data analysis method for flow cytometer detection according to claim 1, wherein The establishment and training steps of the normal signal analysis model specifically include: Establish a training data set and a fine-tuning data set. The training data set is the set of electrical pulse signals collected by using eight-peak rainbow balls for flow cytometer detection as the training electrical pulse signal set. The data after preprocessing the training electrical pulse signal set, including denoising, filtering, and enhancement, is used as the input for training, and the output of training is the size of the eight-peak rainbow balls; The fine-tuning data set is the set of electrical pulse signals obtained by performing flow cytometry on a small sample of cells of the same type as the sample detected by the flow cytometer as the fine-tuning electrical pulse signal set. The data after preprocessing the fine-tuning electrical pulse signal set, including denoising, filtering, and enhancement, is used as the input for fine-tuning. The output of fine-tuning is the representative cell set of the small sample. The representative cell set includes 10 to 20 cells in the small sample. The representative cell set also includes the ratio of each representative cell. The number of cells in the small sample is M times of 106, where M ranges from 5 to 10; Use a convolutional neural network to establish a prototype of the normal signal analysis model, and use the training data set to train the prototype of the normal signal analysis model to obtain the normal signal analysis model; Fine-tune the normal signal analysis model using the fine-tuning data, and use the fine-tuned normal signal analysis model as the output normal signal analysis model.

5. A method for analyzing flow cytometer detection data according to claim 4, characterized in that, The steps to obtain a representative collection of cells from a small sample and the proportion of each representative cell are as follows: Step 1: Take a small amount of cell sample to be tested, and set the number of cells as K; the value of K is 5 to 10 times the total number of cells to be tested; Step 2: Use a microscope to directly observe the sample, and manually count the number of cells with different morphologies observed. Let the number of different cell types observed be W, where W is 10 to 20; Step 3. For each cell category, count its quantity proportion in the sample. Let the number of the i-th type of cells be w i , then the proportion of the i-th type of cells is p i = w i / K; Step 4. Select M representative cells according to the morphological characteristics of the cells to form a representative cell set R = {r1, r2, …, r} W ; Step 5: Construct the corresponding quantity ratio P = p1, p2, …, p for each type of representative cell W ; Step 6: The representative cell set R and its proportion P obtained through observation and statistics are used as the representative cell set and proportion of the small sample cells.

6. A method for analyzing flow cytometry detection data according to claim 1, wherein The method for classifying and analyzing the abnormal signal set is a decision tree algorithm.

7. A method for analyzing flow cytometry detection data according to any one of claims 1-6, characterized in that, The electrical pulse signal includes time, length, width and area.

8. A method for analyzing data detected by a flow cytometer according to claim 4, characterized in that, The method of fine-tuning the conventional signal analysis model using the fine-tuning data is to fine-tune only the last three layers of the conventional signal analysis model.

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 flow cytometer detection data analysis method according to any one of claims 1 to 8.

10. A flow cytometer detection data analysis system, characterized in that, A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images