Method and marker for identifying cell population passage number

Through SR-FTIR technology and machine learning methods, a cell generation identification model is constructed, which solves the problem of cell population generation identification in the existing technology, and achieves rapid and accurate cell quality evaluation and supports the quality control of cell therapy.

WO2024240198A1PCT designated stage expired Publication Date: 2024-11-28SHANGHAI TECH UNIV +2
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2024/094811
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-22
Filing Date
2024-05-22
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

The existing cell quality control methods rely on the experience of experimenters, are highly invasive, and are difficult to quickly and accurately identify cell population generations, and are unable to effectively assist in the quality control of cell preparations, especially in cell therapy, which affects cell activity and safety.

Method used

SR-FTIR technology combined with machine learning methods to build a generation identification model, and the training data set is constructed through the infrared spectral peaks and displacement values ​​of cells. The final generation identification model is trained using GBDT or XGBoost algorithm, combining cluster analysis and decreasing. Dimensional treatment, rapid identification of cell population generation.

Benefits of technology

It has achieved rapid and accurate identification of cell population generation, overcome the shortcomings of traditional methods, provided high-correct cell quality evaluation methods, supported the quality control of cell therapy, and discovered cell proteins, lipid characteristic regions and physiological states. Relevance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2024094811_28112024_PF_FP_ABST
    Figure CN2024094811_28112024_PF_FP_ABST
Patent Text Reader

Abstract

A method for identifying a cell population passage number. The method comprises: judging the probability of a cell population to be identified being in each of N known passage numbers on the basis of a preset passage number identification model; comparing a characteristic quantity of the cell population to be identified with characteristic quantities of cell populations of N known passage numbers to obtain a passage number range within which the cell population to be identified falls; and obtaining a passage number for the cell population to be identified according to the passage number range within which the cell population to be identified falls and the probability of the cell population to be identified being in each passage number. Further disclosed is a marker set for identifying a cell population passage number.
Need to check novelty before this filing date? Find Prior Art

Description

Methods and markers for identifying cell population generations

[0001] Cross-reference to related applications

[0002] This patent application claims priority to Chinese patent application No. 2023105723311, filed on May 22, 2023, entitled “Methods and Markers for Identifying the Generation of Stem Cell Populations”, the full text of which is incorporated herein by reference. Technical Field

[0003] The present invention relates to the field of biomedicine, and in particular to methods and markers for identifying the generations of cell populations including stem cells. Background Art

[0004] Cell therapy requires the use of human-derived live cell preparations, prepared through appropriate in vitro manipulations (such as isolation, culture, expansion, and genetic modification) consistent with approved drug approval. These include genetically modified or non-genetically derived cells, such as autologous or allogeneic immune cells, stem cells, tissue cells, or cell lines. The quality of these human-derived live cell preparations varies, influenced by factors such as their source, type, and characteristics. Due to individual differences in donor material and the type of preparation, production batches can vary significantly.

[0005] When cell preparations are used in clinical cell therapy, the number of passages during the preparation process is a crucial factor influencing cell viability, clinical efficacy, and safety. Biological characteristics of cells of different passages, such as cell morphology, cell markers, karyotype, and cytokine production, vary with culture. Traditional cell quality control methods analyze these biological characteristics individually. For example, morphological examinations are performed using light microscopy; cell viability and cell markers are analyzed using flow cytometry and immunofluorescence staining; and PCR, ELISA, and other experimental methods are used to analyze the levels of specific genes or proteins in cell populations. These existing cell quality control methods not only rely heavily on the experience of the laboratory technician, but some analytical methods are also highly invasive, potentially altering the normal cell growth environment and physiological functions, and even disrupting cell structure, causing irreversible cell damage and necrosis. This makes them unsuitable for rapid testing of large-scale cell samples.

[0006] The emergence of high-throughput sequencing technology has made comprehensive cell characterization possible. However, this technology is complex, with numerous subcategories and varying analytical emphases. Testing is time-consuming, data collection and analysis are complex, and it remains a cross-sectional analysis, unable to reveal information related to cell culture passages and cellular aging trends. Therefore, research and development of a method to objectively identify the passage number of any given batch of cells in a cell preparation is needed to assist in quality control of cell preparations.

[0007] Summary of the Invention

[0008] The object of the present invention is to provide a method for identifying the passage number of a cell population.

[0009] Another object of the present invention is to identify marker panels of cell population passages.

[0010] Another object of the present invention is a method for screening markers that can be used to identify the passage of a cell population.

[0011] To solve the above technical problems, the first aspect of the present invention provides a method for identifying the passage number of a cell population, the method comprising the steps of:

[0012] Determine the probability that the cell population to be identified is at each generation among the cell populations of N known generations based on a preset generation identification model;

[0013] Comparing the characteristic values ​​of the cell population to be identified with the characteristic values ​​of N cell populations of known generations to obtain the generation range to which the cell population to be identified belongs;

[0014] The generation of the cell population to be identified is obtained according to the generation range to which the cell population to be identified belongs and the probability that the cell population to be identified is in each generation.

[0015] In some preferred embodiments, the cells are stem cells.

[0016] In some preferred embodiments, the value of N is an integer not less than 3, such as 3 or 4.

[0017] In some preferred embodiments, the preset generation identification model is constructed by the following method:

[0018] Constructing a training data set, comprising the characteristic quantities of the N cell populations with known passage numbers and the corresponding passage number values, wherein the characteristic quantities of the cell populations include several characteristic peak values ​​and displacement values ​​obtained by SR-FTIR analysis of the cells;

[0019] constructing an initial identification model, wherein the initial identification model includes an independent variable and a dependent variable, wherein the independent variable is a characteristic value of the cell population and the dependent variable is the generation number of the cell population;

[0020] The initial identification model is trained based on the training data set to obtain a final generation identification model.

[0021] In some preferred embodiments, the initial identification model is trained using the GBDT algorithm to obtain a final generation identification model.

[0022] In some preferred embodiments, the characteristic quantities of the cells include four characteristic peak values ​​and corresponding displacement values ​​of νas(CH3), NH bending of Amide I, α-helical of Amide I and α-helical of Amide II obtained by SR-FTIR analysis of the cells.

[0023] In some preferred embodiments, the characteristic quantities of the cells further include νas(CH2), νs(CH3), νs(CH2), β-pleated sheet of Amide I, β-pleated sheet of Amide II, "Tyrosine" band, νas(PO 2- ),νs(PO 2- ), characteristic peak values ​​and corresponding displacement values ​​of CO stretching vibratios:DNA and CO stretching vibratios:RNA.

[0024] In some preferred schemes, the characteristic quantities of the cell population to be identified and the characteristic quantities of N cell populations of known generations are subjected to dimensionality reduction processing, and then the two are compared to obtain the generation range to which the cell population to be identified belongs.

[0025] In some preferred embodiments, the dimensionality reduction process causes the characteristic quantities of the cell population to be identified and the characteristic quantities of the N known cell populations to be in the same dimension.

[0026] In some preferred embodiments, cluster analysis is performed on the characteristic quantities of the cell population to be identified after dimensionality reduction processing to obtain non-outlier points of the cell population to be identified; and

[0027] performing cluster analysis on the characteristic quantities of the N cell populations of known generations that have undergone dimensionality reduction processing, respectively, to obtain cluster centers of the N cell populations of known generations;

[0028] By comparing the distances between the non-outlier points of the cell population to be identified and the cluster centers of the N cell populations of known generations, the generation range to which the cell population to be identified belongs is obtained.

[0029] In some preferred schemes, by comparing the distances between the non-outlier points of the cell population to be identified and the cluster centers of the N cell populations of known generations, at least two target cluster centers are selected from the cluster centers of the N cell populations of known generations, and the generation range is composed according to the generation of the cell population of known generations to which the target cluster centers belong.

[0030] In some preferred embodiments, a first target cluster center that is closest to a non-outlier point of the cell population to be identified is selected from the cluster centers of the N known cell populations;

[0031] Selecting cluster centers of two cell populations of known generations adjacent to the generation to which the first target cluster center belongs, comparing the distances between the cluster centers of the two adjacent cell populations of known generations and the non-outlier points of the cell population to be identified, and selecting the cluster center of the cell population of known generation that is closer to the non-outlier points of the cell population to be identified as the second target cluster center;

[0032] The generation range is composed of the generation of the cell population of known generation to which the first target cluster center and the second target cluster center belong.

[0033] In some preferred schemes, when the cluster centers of two cell populations of known generations adjacent to the generation to which the first target cluster center belongs are at an equal distance from the non-outlier points of the cell population to be identified, the generations of the cell population of known generations to which the cluster centers of the two cell populations of known generations adjacent to the generation to which the first target cluster center belongs are used to form the generation range.

[0034] In some preferred schemes, a first target cluster center that is closest to the non-outlier point of the cell population to be identified and a second target cluster center that is second closest to the cluster center of the cell population to be identified are selected from the cluster centers of the N cell populations of known generations; and the generations of the cell populations of known generations to which the first target cluster center and the second target cluster center belong are used to form the generation range.

[0035] In some preferred embodiments, the cluster analysis is any one of DBCAN cluster analysis and K-means cluster analysis or a combination of the two.

[0036] In some preferred embodiments, K-means and / or DBSCAN cluster analysis is performed on the characteristic quantities of the N known generations of cell populations that have undergone dimensionality reduction processing and / or the characteristic quantities of the cell population to be identified.

[0037] In some preferred embodiments, the dimensionality reduction process is PCA dimensionality reduction process.

[0038] In some preferred embodiments, a weighted average formula is used to calculate the generation range of the cell population to be identified and the probability that the cell population to be identified is in each generation to obtain the generation of the cell population to be identified.

[0039] In some preferred solutions, the weighted average formula is as follows:

[0040] The passage number of the cell population to be identified = (ProbP X1 ×X1+ProbP X2 ×X2+……+ProbP Xn ×Xn) / (ProbP X1 +ProbP X2 +……+ProbP Xn );

[0041] Among them, ProbP X1 The cell population to be identified is at passage P X1 The probability of

[0042] Prob P X2 The cell population to be identified is at passage P X2 The probability of

[0043] Prob P Xn The cell population to be identified is at passage P Xn probability.

[0044] In some preferred embodiments, the probability of the generation of the cell population to be identified is

[0045] =1-ProbP Xn -ProbP Xn -1-……-ProbP X1 ×ProbP X2 / 2,

[0046] Among them, ProbP X1 The cell population to be identified is at passage P X1 The probability of

[0047] Prob P X2 The cell population to be identified is at passage P X2 The probability of

[0048] Prob P Xn The cell population to be identified is at passage P Xn probability.

[0049] In some preferred solutions, the weighted average formula is as follows:

[0050] The passage number of the cell population to be identified = ProbP X1 ×X1+ProbP X2 ×X2) / (ProbP X1 +ProbP X2 );

[0051] Among them, ProbP X1 The cell population to be identified is at passage PX1 The probability of

[0052] Prob P X2 The cell population to be identified is at passage P X2 The probability of

[0053] P X1 is the value of the X1th generation,

[0054] P X2 is the value of the X2th generation,

[0055] The generation range of the cell population to be identified is P X1 -P X2 .

[0056] In some preferred embodiments, the probability formula for identifying the cell population generation is as follows:

[0057] The probability that the passage of the cell population to be identified is obtained by the method

[0058] =1-ProbP Xn -ProbP Xn -1-……-ProbP X1 *×ProbP X2 / 2;

[0059] Among them, ProbP X1 The cell population to be identified is at passage P X1 The probability of

[0060] Prob P X2 The cell population to be identified is at passage P X2 The probability of

[0061] P X1 is the value of the X1th generation,

[0062] P X2 is the value of the X2th generation,

[0063] The generation range of the cell population to be identified is P X1 -P X2 .

[0064] In a second aspect of the present invention, a marker panel for identifying the passage of a cell population including stem cells is provided, wherein the marker panel comprises the following characteristic peaks: νas(CH3), NH bending of Amide I, α-helical of Amide I and α-helical of Amide II.

[0065] In some preferred embodiments, the marker group further includes any one of the following characteristic peaks: νas(CH2), νs(CH3), νs(CH2), β-pleated sheet of Amide I, β-pleated sheet of Amide II, "Tyrosine" band, νas(PO 2- ),νs(PO 2- ), CO stretching vibrations: DNA and CO stretching vibrations: RNA.

[0066] A third aspect of the present invention provides a method for identifying the passage number of a cell population, comprising the steps of:

[0067] A cell to be identified is extracted from the cell population to be identified as a target cell, and N cells are randomly extracted from the cell population to be identified (preferably, less than 10 cells are randomly extracted, preferably not less than 50 cells, more preferably not less than 100 cells are randomly extracted) to form a first cell cluster with the target cell,

[0068] Obtaining characteristic quantities of cells in the first cell cluster, and using a preset generation identification model to make a determination to obtain a first identification generation value corresponding to the target cell;

[0069] Again, N cells are randomly extracted from the identified cell group to form a second cell cluster with the target cells.

[0070] Obtaining a characteristic value of the second cell cluster, and using a preset generation identification model to determine the second identification generation value corresponding to the target cell;

[0071] This cycle is repeated to obtain multiple identification generation values ​​corresponding to the target cells, and the generation of the cell population to be identified is obtained by performing data processing on the multiple predicted generation values.

[0072] In some preferred embodiments, the characteristic value of the cell cluster is obtained, and data processing is performed on it (preferably, the data processing is averaging, median or standard deviation processing), and then a preset generation model is used for judgment.

[0073] In some preferred embodiments, the data processing is to obtain the median or average of the identification generation values ​​of the obtained cells.

[0074] In some preferred embodiments, the preset generation identification model is constructed by the following method:

[0075] Constructing a training data set, wherein the training data set includes characteristic quantities of N cell populations with known passage numbers and corresponding passage number values;

[0076] constructing an initial identification model, wherein the initial identification model includes an independent variable and a dependent variable, wherein the independent variable is a characteristic value of the cell population and the dependent variable is the generation number of the cell population;

[0077] Training the initial identification model based on the training data set to obtain a final generation identification model;

[0078] The characteristic quantities of the cells include several characteristic peak values ​​and displacement values ​​obtained by SR-FTIR analysis of the cells.

[0079] In some preferred embodiments, the initial identification model is trained using the XGBoost algorithm to obtain a final generation identification model.

[0080] In some preferred embodiments, the characteristic peaks are screened based on peak parameters obtained from SR-FTIR analysis, wherein the peak parameters include peak number, peak order, peak wave number and / or peak morphology.

[0081] In some preferred schemes, the method for screening the characteristic peak includes the steps of: determining a preset range of displacement values ​​of the target characteristic peak, screening according to the number of peaks, peak order, peak displacement values ​​and / or peak morphology within the preset range, and selecting a peak that matches the number of peaks, peak order, peak displacement values ​​and / or peak morphology within the preset range of the target characteristic peak as the characteristic peak.

[0082] In some preferred embodiments, the characteristic peaks are screened using the XGBoost algorithm.

[0083] In some preferred embodiments, the characteristic quantities of the cells include Amide A, Amide B, Olefinic=CH stretching, Vas(CH3), Vas(CH2), Vs(CH3), Vs(CH2), C=O stretching of lipid esters, C=O antisymmetric stretching: RNA and purine base, NH bending of Amide I, a-helical of Amide I, b-pleated sheet of Amide I, C4-C5 and C=N stretching in imidazole ring of DNA, C4-C5 and C=N stretching in imidazole ring of RNA, a-helical of Amide II, b-pleated sheet of Amide II, Tyrosine, CH2 bending vibrations, CH3 bending and CH2 scissoring vibrations, COO-symmetric stretching, Peptide side chain vibrations, Vas(PO2-), CO–O–C antisymmetric stretching vibrations of glycogen and nucleic acid obtained by SR-FTIR analysis of the cells. Peak values ​​and corresponding displacement values ​​of ribose, Vs(PO2-), CO stretching vibratios:DNA, CO stretching vibratios:RNA and / or Mainly from glycogen characteristic peaks.

[0084] In a fourth aspect of the present invention, a marker panel for identifying the passage of a cell population including stem cells is provided, wherein the marker panel comprises the following characteristic peaks: Amide A, Amide B, Olefinic=CH stretching, Vas(CH3), Vas(CH2), Vs(CH3), Vs(CH2), C=O stretching of lipid esters, C=O antisymmetric stretching: RNA and purine base, NH bending of Amide I, a-helical of Amide I, b-pleated sheet of Amide I, C4-C5 and C=N stretching in imidazole ring of DNA, C4-C5 and C=N stretching in imidazole ring of RNA, a-helical of Amide II, b-pleated sheet of Amide II, Tyrosine, CH2 bending vibrations, CH3 bending and CH2 scissoring vibrations, COO-symmetric stretching, Peptide side chain vibrations, Vas(PO2-), CO–O–C antisymmetric stretching vibrations of glycogen and nucleic acid ribose, Vs(PO2-), CO stretching vibratios:DNA, CO stretching vibratios:RNA and Mainly from glycogen.

[0085] Compared with the prior art, the present invention has at least the following advantages:

[0086] (1) The method for identifying cell population generations provided by the present invention innovatively employs SR-FTIR technology. Furthermore, by establishing a relevant mathematical model, a correlation between the infrared spectrum of cells and their physiological state is established. This can be used to rapidly obtain standardized information on cell quality, providing a quality evaluation method required for clinical application in cell therapy for related diseases.

[0087] (2) SR-FTIR technology has high spatial resolution within the diffraction limit and high sensitivity in identifying intercellular variations. The cell population generation identification method of the present invention overcomes the shortcomings of traditional biological techniques and can identify the relative physiological generation of cells with a high accuracy rate;

[0088] (3) The present invention also discovered the correlation between the SR-FTIR spectra of the characteristic regions corresponding to cell proteins and lipids and the physiological state of cells, which provides a basis and guarantee for the subsequent development of a cell characteristic spectrometer that can quickly collect SR-FTIR for the evaluation of cell physiological state;

[0089] (4) The present invention also clarifies that a corresponding cell generation identification model needs to be established to identify cells from different sources.

[0090] It should be understood that within the scope of the present invention, the above-mentioned technical features of the present invention and the technical features described in detail below (such as in the embodiments) can be combined with each other to form new or preferred technical solutions. Due to space limitations, they will not be listed here one by one. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] One or more embodiments are exemplarily described by the figures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments.

[0092] FIG1 is a schematic diagram of a single-cell SR-FTIR spectral data preprocessing process according to an embodiment of the present invention;

[0093] FIG2 is a schematic diagram of a cell generation identification model process according to an embodiment of the present invention;

[0094] Figure 3 is a histogram and scatter plot of the modeling results of Model A, Model B and Model C according to an embodiment of the present invention, A: histogram and scatter plot of the modeling results of Model A, B: histogram and scatter plot of the modeling results of Model B, C: histogram and scatter plot of the modeling results of Model C, the histogram results of the model determination generation distribution are fitted using Gaussian, and the median and 95% confidence interval are marked in the scatter plot of the generation and accuracy of each single cell model;

[0095] Figure 4 is a histogram and scatter plot of the model validation results of Model A, Model B and Model C according to an embodiment of the present invention, A: Model A model validation result histogram and scatter plot, from left to right are SIAISi011-A P10, P20, P30; B: Model B model validation result histogram and scatter plot, from left to right are SIAISi011-A P10, P20, P30, P35; C: Model C model validation result histogram and scatter plot, from left to right are SIAISi011-A P5, P10, P20, P30, the model determination generation distribution histogram results adopt Gaussian fitting, and the median and 95% confidence interval are marked in the scatter plot of each single cell model determination generation and accuracy;

[0096] FIG5 is a histogram and scatter plot of the modeling results of Model AT, Model BT and Model CT according to an embodiment of the present invention, A: histogram and scatter plot of the modeling results of Model AT, B: histogram and scatter plot of the modeling results of Model BT, C: histogram and scatter plot of the modeling results of Model CT, the histogram results of the model determination generation distribution are fitted using Gaussian, and the median and 95% confidence interval are marked in the scatter plot of the generation and accuracy of each single cell model;

[0097] Figure 6 is a histogram and scatter plot of the model validation results of Model AT, Model BT and Model CT according to an embodiment of the present invention, A: Model AT model validation result histogram and scatter plot, from left to right are TUSMi002 P10, P20, P30; B: Model BT model validation result histogram and scatter plot, from left to right are TUSMi002 P10, P20, P30, P35; C: Model CT model validation result histogram and scatter plot, from left to right are TUSMi002 P5, P10, P20, P30, the model determination generation distribution histogram results adopt Gaussian fitting, and the median and 95% confidence interval are marked in the scatter plot of each single-cell model determination generation and accuracy;

[0098] Figure 7 shows simplified modeling results according to an embodiment of the present invention: A: SIAISi011-A stem cell modeling results; B: TUSMi002 stem cell modeling results. Gaussian fit was used to represent the distribution histogram of model-determined cell generations; the median and 95% confidence interval are plotted in the scatter plots of each cell model's generation and accuracy.

[0099] Figure 8 shows the results of the simplified model for the identification group according to an embodiment of the present invention. A: The results of the simplified model SIAISi011-A stem cell identification group, with the generations from left to right being P10, P20, and P30 respectively; B: The results of the simplified model TUSMi002 stem cell identification group, with the generations from left to right being P10, P20, and P30 respectively; the histogram of the generation distribution of the model determination was fitted using Gaussian fitting; the median and 95% confidence interval are marked in the scatter plots of the generation determination and accuracy of each single cell model;

[0100] Figure 9 shows the cross-validation model validation results using Model A and Model AT modeling groups according to an embodiment of the present invention. A and B are the histograms and scatter plots of the validation results of Model A for TUSMi002 stem cells at P5, P15, P25, and P35, respectively; and E and F are the histograms and scatter plots of the validation results of Model AT for SIAISi011-A stem cells at P5, P15, P25, and P35, respectively. The histograms of the model generation distribution were fitted using a Gaussian fit. The median and 95% confidence interval are marked in the scatter plots of the model generation and accuracy for each single-cell model.

[0101] Figure 10 shows the determination results of the test group by cross-validation using Model A and Model AT modeling groups according to an embodiment of the present invention. AD are the histogram and scatter plot of the determination results of Model A on TUSMi002 stem cells P10, P20, and P30, respectively; EF are the histogram and scatter plot of the determination results of Model AT on SIAISi011-A stem cells P10, P20, and P30, respectively. The model determination generation distribution histogram results are fitted with Gaussian fitting; the median and 95% confidence interval are marked in the scatter plot of the determination generation and accuracy of each single cell model.

[0102] FIG11 is a flow chart of modeling another cell generation identification model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0103] The present invention has developed a method for identifying the generation of a cell population including stem cells based on machine learning. This method can identify the generation of a given batch of cells and can be used to assist in controlling cell quality.

[0104] Models and methods for characterizing cell population passages

[0105] An embodiment of the present invention relates to a method for identifying the passage of a cell population, specifically comprising steps S1-S3.

[0106] S1, based on a preset generation identification model, determining the probability that the cell population to be identified is in each of N known generations.

[0107] The method for constructing the preset generation identification model used in the above step S1 includes the following sub-steps S1.1-S1.3.

[0108] In a preferred embodiment, the cells are stem cells.

[0109] S1.1 constructs a training data set, which is the characteristic quantity information of a cell group with a known number of passages measured in advance. The characteristic quantity information of these cell groups is obtained by taking N cell groups with a known number of passages for SR-FTIR analysis, and the peak values ​​and corresponding displacement values ​​of the characteristic peaks are obtained. The N value here is not less than 2, but preferably, not less than 3, for example, it can be 3, 4, 5, or 6. For example, the characteristic quantities of cell groups of four preset generations P5, P10, P15, and P20 can be used to form a training data set, or the characteristic quantities of cell groups of six preset generations P5, P10, P15, P20, P25, and P30 can be used to form a training data set. It should be understood that the cell group generations and the number of cell groups selected in the training data set are not limited thereto.

[0110] In some embodiments, the characteristic quantities of the cell population include at least four characteristic peak values ​​and corresponding displacement values ​​of νas(CH3), NH bending of Amide I, α-helical of Amide I and α-helical of Amide II obtained by SR-FTIR analysis of the cell population.

[0111] S1.2 constructs an initial identification model, using characteristic quantities of the cell population, such as the peak value and corresponding displacement value of the characteristic peak of the cell population as independent variables, and the corresponding number of cell population passages as the dependent variable to construct the initial identification model.

[0112] S1.3 trains the initial identification model based on the training data set to obtain the final generation identification model.

[0113] In step S1.3, the initial identification model is trained using the GBDT algorithm to obtain the final generation identification model. "GBDT" (Gradient Boosting Decision Tree) refers to the gradient descent tree. "Gradient Boosting" is an algorithm within the ensemble boosting method. It iterates new learners using gradient descent. It is an iterative decision tree algorithm composed of multiple decision trees, and the conclusions of all trees are accumulated to form the final answer.

[0114] The generation of a cell population can also be identified using only the above-mentioned generation identification model. However, the inventors found that the generation identification model obtained by a single GBDT algorithm has redundant information interference in the identification of generations, resulting in a low model accuracy. For example, when a cell population of unknown generation is obtained, using a single GBDT algorithm will calculate the probability of the cell population to be identified at each generation separately and take a weighted average. In this way, the probabilities of many preset generations that have a poor correlation with the cell population to be identified will be included in the algorithm, resulting in a decrease in accuracy.

[0115] For example, when the model is trained using data from cell populations of four preset generations, P5, P10, P15, and P20, a generation identification model A1 is obtained. When identification is performed using the generation identification model A1, the probabilities of the cell population to be identified being at the four preset generations, P5, P10, P15, and P20, are obtained, respectively, and are recorded as ProbP5, ProbP10, ProbP15, and ProbP20. When the above probabilities are directly used for weighted averaging, the cell generation obtained by identification can be calculated using the following formula I:

[0116] The passage number of the cell population to be identified = (ProbP5×5+ProbP10×10+ProbP15×15+ProbP20×20) / (5+10+15+20) Formula I.

[0117] However, among the four preset passages P5, P10, P15, and P20, some of the cell populations at these preset passages have a low correlation with the cell population to be identified. If the probability data of these preset passages with low correlation are included in the formula for weighted average, the result obtained will inevitably differ greatly from the actual passage number of the cell population to be identified. Therefore, the inventors optimized the model and added the passages P corresponding to the two preset passages that are most relevant to the true characteristics of the cell population to be identified. x1 and P x2 The actual number of passages of the cells to be identified should be P x1 -P x2 In this case, only the two most relevant generations P need to be used. x1 and P x2 The generation P of the cell populations with the two preset generations that are most relevant to the characteristics of the cell population to be identified is calculated by excluding other items with low correlation from the calculation formula. x1 and P x2 , and the passage number range described in the actual passage number of the cells to be identified can be obtained by the following step S2.

[0118] S2, comparing the characteristic values ​​of the cell population to be identified with the characteristic values ​​of N cell populations of known generations, to obtain the generation range to which the cell population to be identified belongs.

[0119] In step S2, preferably, to eliminate data with low impact factors and facilitate comparison, the characteristic values ​​of the cell population to be identified and the characteristic values ​​of the N known cell populations can be subjected to dimensionality reduction processing to bring them to the same dimension. The two can then be compared to obtain the generation range to which the cell population to be identified belongs. Preferably, the dimensionality reduction processing is PCA dimensionality reduction processing.

[0120] The method for comparing the characteristic values ​​of the cell population to be identified with the characteristic values ​​of N cell populations of known passages may include the following methods:

[0121] Perform K-means and / or DBSCAN cluster analysis on the characteristic quantities of the N cell populations of known generations after dimensionality reduction and the characteristic quantities of the cell population to be identified; and obtain the generation range P to which the cell population to be identified belongs by comparing the distances between the non-outlier points of the cell population to be identified and the cluster centers of the N cell populations of known generations. x1 -P x2 .

[0122] In a preferred embodiment, the distance between the non-outlier point of the cell population to be identified and the cluster center of the cell population of N known generations is compared. Among the cluster centers of the cell population of N known generations, the cluster centers of the cell population of at least two known generations close to the non-outlier point of the cell population to be identified are selected as the target cluster center. In one example, the distance between the cluster centers of the cell populations of 1 to N known generations and the non-outlier point of the cell population to be identified is recorded as d1, d2 ... dn (wherein, d1, d2 ... each d value in dn is unequal), now the cluster center of the cell population of the known generation with the smallest d value is selected as the first target cluster center, and the cluster center of the cell population of the known generation with the second smallest d value (second smallest) is selected as the second target cluster center. Then, the generations of the cell population of the known generations to which the first target cluster center and the second target cluster center belong can be used to determine the generation range of the cell population to be identified. Assuming that the smallest d value is d1, the cluster center of the first known generation cell population is the first target cluster center. Assuming that the second smallest d value is d2, the cluster center of the second known generation cell population is the second target cluster center. Then, the generation range of the cell population to be identified is between the generation corresponding to the cluster center of the first known generation cell population and the generation corresponding to the cluster center of the second known generation cell population. In the present invention, "second" and "second" refer to the value next to the maximum value.

[0123] In one example, the distances between the cluster centers of 1 to N known generations of cell populations and the non-outlier points of the cell population to be identified are recorded as d1, d2...dn, (wherein, each d value in d1, d2...dn may be equal). In this case, the cluster center of the cell population of known generations corresponding to the smallest d value may not be unique. For example, assuming that the smallest d value is d1, d2, and d3, and d1=d2=d3, or the cluster center of the cell population of known generations corresponding to the smallest d value is unique, the cluster center corresponding to the next smallest d value may not be unique. For example, assuming that the smallest d value is d1, the next smallest d value is d2 and d3, and d2=d3, then the minimum range of generations that can include the cluster centers of these two or three known cell populations is determined as the range of generations of the cell population to be identified.

[0124] Once the passage numbers P corresponding to the two passages of cells that are most relevant to the characteristics of the cell population to be identified are obtained, x1 and P x2 , the passage number of the cell population can be identified by the following step S3.

[0125] S3, obtaining the generation of the cell population to be identified according to the generation range to which the cell population to be identified belongs and the probability that the cell population to be identified is in each generation.

[0126] Specifically, the generation number of the cell population to be identified can be obtained by the following formula II:

[0127] The passage number of the cell population to be identified = (ProbP X1 ×X1+ProbP X2 ×X2+…+ProbP Xn ×Xn) / (ProbP X1 +ProbP X2 +……+ProbP Xn ); Formula II.

[0128] Among them, ProbP X1 The cell population to be identified is at passage P X1 The probability of

[0129] ProbP X2 The cell population to be identified is at passage P X2 The probability of

[0130] ProbP Xn The cell population to be identified is at passage P Xn probability.

[0131] The generation of the cell population to be identified can be obtained by substituting the generation values ​​of the cluster centers of at least two known cell populations close to the non-outlier points of the cell population to be identified obtained in step S2 into the above formula II.

[0132] Generally, the clustering center of the cell mass of the known generation that is close to the non-outlier point of cell mass to be identified appears in the clustering center of the cell mass of two adjacent generations to which the clustering center of the cell mass of the known generation that is closest to the non-outlier point of cell mass to be identified belongs.For example, when using the cell data training model of four preset generations of P5, P10, P15, P20, when the clustering center to which cell mass to be identified belongs and the clustering center of P15 for cell mass approached, the generation that usually belongs to the clustering center closest to the non-outlier point of cell mass to be identified is at least one in the P10 and P20 generations, or can be 2.So now, can by comparing the distance of the clustering center of P10 and P20 cell mass and the non-outlier point of cell mass to be identified respectively which is relatively closer, directly determine the generation scope to which cell mass to be identified belongs. That is, by comparing the distances between the cluster centers of two adjacent generations of cell populations and the non-outlier point of the cell population to be identified, the generation range of the cell population to be identified is determined.

[0133] In some examples, when the number of cluster centers of cell populations of known generations closest to the non-outlier points of the cell population to be identified is greater than one, for example, two or three, the two generation values ​​P of the minimum generation range of cell populations including these two or three known generations and the probability ProbP that the cell population to be identified is in these two generations are substituted into Formula II to obtain the generation of the cell population to be identified and the probability of identification.

[0134] In some examples, the number of cluster centers of cell populations of known generations closest to the non-outlier points of the cell population to be identified is one, but the number of cluster centers of cell populations of known generations 20 times closer to the non-outlier points of the cell population to be identified is greater than 1, for example, 2 or 3. In this case, the generation value P corresponding to the cluster center of the cell population of known generations closest to the non-outlier points of the cell population to be identified and the probability Prob P that the cell population to be identified is at this generation are substituted into Formula II, and the generation values ​​P corresponding to 2 or 3 cluster centers of cell populations of known generations 20 times closer to the non-outlier points of the cell population to be identified and the probability Prob P that the cell population to be identified is at this generation are substituted into Formula II to obtain the generation of the cell population to be identified.

[0135] In some examples, the number of cluster centers of cell populations of known generations closest to the cluster center of the cell population to be identified is one, and the number of cluster centers of cell populations of known generations second closest to the cluster center of the cell population to be identified is one. In this case, the generation of the cell population to be identified can be calculated using the following formula III.

[0136] The passage number of the cell population to be identified = Prob P X1 ×P X1+Prob P X2 ×P X2 ) / (Prob P X1 +Prob P X2 ) Formula III.

[0137] In Formula III, Prob P X1 is the probability that the cell population to be identified is at the X1th generation,

[0138] Prob P X2 is the probability that the cell population to be identified is at the X2 generation,

[0139] P X1 is the value of the X1th generation,

[0140] P X2 is the value of the X2th generation,

[0141] X1≠X2,

[0142] The generation range of the cell population to be identified is P X1 To P X2 .

[0143] On the other hand, the probability that the generation of the cell population to be identified is the generation calculated by formula II can be obtained by the following formula III:

[0144] Probability = 1-ProbP Xn -ProbP Xn-1 -……-ProbP X1 ×ProbP X2 / 2; Formula III.

[0145] Marker panel

[0146] Embodiments of the present invention also relate to a marker panel for identifying passages of a cell population comprising stem cells, the marker panel comprising the following characteristic peaks: νas(CH3), NH bending of Amide I, α-helical of Amide I, and α-helical of Amide II.

[0147] In the present invention, the peak values ​​of the marker characteristic peaks of the control cell population and the cell population of the generation to be identified can be collected by using SR-FTIR spectral analysis, and the cell population generation can be identified by comparing the peak values ​​and displacements of the marker characteristic peaks of the control cell population and the cell population of the generation to be identified; wherein, the marker characteristic peaks include at least νas(CH3), NH bending of Amide I, α-helical of Amide I and α-helical of Amide II. Preferably, the marker group also includes at least one characteristic peak selected from the following: νas(CH2), νs(CH3), νs(CH2), β-pleated sheet of Amide I, β-pleated sheet of Amide II, "Tyrosine" band, νas(PO 2- ),νs(PO 2- ), CO stretching vibrations: DNA and CO stretching vibrations: RNA.

[0148] Self-model building group

[0149] Embodiments of the present invention also involve mutual verification of cell generation identification models established for cells from different sources. The inventors discovered that to identify the generation of cells from any source, it is necessary to collect SR-FTIR spectra of each corresponding cell population generation. The present invention's cell population generation identification model construction method was used to establish a self-specific cell population generation identification model.

[0150] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the present invention will be further described below in conjunction with specific examples. It should be understood that these examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention. The experimental methods in the following examples, for which specific conditions are not specified, are usually based on conventional conditions or the conditions recommended by the manufacturer. Unless otherwise stated, percentages and parts are weight percentages and parts by weight. The experimental materials and reagents used in the following examples can be obtained from commercial sources unless otherwise specified.

[0151] Unless otherwise specified, the technical and scientific terms used herein have the same meaning as commonly understood by ordinary technicians in the technical field to which this application belongs. It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments of this application.

[0152] Example 1

[0153] In this example, a set of reliable SR-FTIR spectral data from different cell generations was collected, demonstrating that SR-FTIR can identify differences in biochemical composition or structure between cells of the same source and different cell generations. The specific implementation steps are as follows:

[0154] Two stem cell lines derived from healthy individuals (SIAISi011-A, TUSMi002) were selected, and different cell passages of serially passaged cells (P5±2, P10±2, P15±2, P20±2, P25±2, P30±2, P35±2) were selected. Hundreds of cells of each passage were seeded on calcium fluoride glass slides covered with matrigel, cultured for 24 hours, fixed, and completely dried at room temperature. SR-FTIR spectra were collected at the BL01B beamline of the Shanghai Light Source Center. For each sample, single-cell spectra of at least 60 cells were randomly collected, with the aperture setting equivalent to the diameter of a single cell, and the spectral range of 4000-900cm -1 , scan number 64, spectral resolution 4cm -1 All SR-FTIR data were automatically baselined and smoothed with SG-15 points, then Mie scattering correction was performed using the RmieS-EMSC correction. Finally, the second-order derivative spectra were obtained using the Savitzky-Golay algorithm. The data processing flow chart is shown in Figure 1.

[0155] Different wavenumber ranges of the second-order derivative spectrum characterize the chemical bonds of different types of substances. There are a series of known specific chemical bond characteristic peaks, and the ranges of each representative characteristic peak are shown in Table 1.

[0156] Table 1

[0157] In Table 1 above, νas represents asymmetric stretching vibration, and vs represents symmetric stretching vibration;

[0158] νas(CH3) represents the asymmetric stretching vibration of methyl group;

[0159] νas(CH2) represents the asymmetric stretching vibration of methylene;

[0160] νs(CH3) represents the asymmetric stretching vibration of methyl group;

[0161] νs(CH2) represents the asymmetric stretching vibration of methylene;

[0162] NH bending of Amide I represents the NH bending vibration of the amide I band;

[0163] α-helical of Amide I indicates the α-helix of the amide I band;

[0164] β-pleated sheet of Amide I indicates the β-pleated sheet of Amide I ribbon;

[0165] α-helical of Amide II indicates the α-helix of the amide II ribbon;

[0166] β-pleated sheet of Amide II indicates the β-pleated sheet of Amide II ribbon;

[0167] “Tyrosine” band indicates tyrosine band;

[0168] νas(PO 2- ) means (PO 2- )'s asymmetric stretching vibration;

[0169] νs(PO 2- ) means (PO 2- )'s symmetrical stretching vibration;

[0170] CO stretching vibratios:DNA represents the stretching vibrations of the CO bonds of DNA;

[0171] CO stretching vibratios:RNA represents the CO bond stretching vibrations of RNA.

[0172] The peak value and displacement of the characteristic peaks characterize the content and strength of the chemical bonds related to each single cell. The higher the peak value, the more the content of the chemical bond in the cell. The shift to higher wavenumbers indicates that the strength of the chemical bond in the cell increases. The characteristic peaks νas(CH3),νas(CH2),νs(CH3),νs(CH2),NH bending of Amide I,α-helical of Amide I,β-pleated sheet of Amide I,α-helical of Amide II,β-pleated sheet of Amide II"Tyrosine" band,C4-C5 and C=N stretching in imidazole ring of RNA,νas(PO 2- ),νs(PO 2- ), CO stretching vibratios:DNA, CO stretching vibratios:RNA) displacement and peak data, and variance analysis was performed (Tables 2-1 to 2-4), which proved that SR-FTIR can identify the differences in biochemical composition or structure of cells of different generations from the same source.

[0173] Table 2-1 Statistics of characteristic peak displacement of SIAISi011-A

[0174] Table 2-2 SIAISi011-A characteristic peak-to-peak statistics

[0175] Table 2-3 Statistics of characteristic peak displacement of TUSMi002

[0176] Table 2-4 TUSMi002 characteristic peak-to-peak statistics

[0177] Example 2

[0178] In this embodiment, a data classification model is constructed to establish an identification system that can be used to evaluate the relative physiological state of cells, and the relative cell generation and identification probability of single cells are confirmed through data weighted analysis.

[0179] (1) Construction of cell generation identification model and cell generation identification method

[0180] To construct a cell generation identification model, it is first necessary to collect several sets of cell data sets of known generations for comparison. This information consists of characteristic peak-to-peak value and displacement data collected using the method described in Example 1. This data constitutes the training data set. The generation identification model is constructed using the GBDT classification model. The initial generation identification model is trained using the training data set to obtain the final cell generation identification model.

[0181] To improve the accuracy of the model, the inventors established another method to provide additional information for classification, which is referred to as the "feature extraction and comparison method".

[0182] The specific steps of the "feature extraction and comparison method" are as follows: using PCA dimensionality reduction to eliminate redundant information in the cells to be identified, and then clustering with DBSCAN to remove outlier cell data; using PCA dimensionality reduction to reduce the dimensionality of the known generation cells in the training data set to the same dimension as the cells to be identified, clustering with kmeans to find the core features (cluster centers) of these groups of cell data, and comparing the features of the cells to be identified with the known generation cell data to fuzzy infer the generation of the current cell data. Combining the results of this fuzzy inference with the judgment results of GBDT, a weighted average is applied to more accurately infer the generation data and its probability for this batch of cells (see Figure 2 for the specific process).

[0183] The model was built using the SIAISi011-A data. The P5, P15, P25, and P35 data sets were selected as pre-input data sets of known cell generations for comparison, hereafter referred to as the training data set. Depending on the number of pre-input data sets, two models were generated: a four-point data set (pre-input P5, P15, P25, and P35, referred to as Model A) and a three-point data set (pre-input P5, P15, P25, referred to as Model B) and a three-point data set (pre-input P15, P15, P25, referred to as Model C). When new P15 cell data (not used in modeling) were input, the GBDT classification model calculated the similarity between the re-input P15 cell data and the data sets in the known modeling data sets. Combined with the cell generation range determined by the "feature extraction and comparison method" described above, the new P15 cell generation was accurately identified. Similarly, accurate generation information was obtained when other cell generation data sets (such as P5, P25, and P35) were input. The results are shown in Figure 3 and Table 3.

[0184] Table 3: Model judgment results of SIAISi011-A model modeling group

[0185] Note: Model A: SIAISi011-A P5-15-25-35; Model B: SIAISi011-A P5-15-25; Model C: SIAISi011-A P15-25-35

[0186] (2) Accuracy verification

[0187] In order to verify the accuracy of the above-mentioned Model A, Model B and Model C, other generations of SIAISi011-A, such as, were used as re-input cell datasets to verify the models. The results are shown in Figure 4 and Table 4.

[0188] Table 4: SIAISi011-A model validation set model judgment results

[0189] Note: Model A: SIAISi011-A P5-15-25-35; Model B: SIAISi011-A P5-15-25; Model C: SIAISi011-AP15-25-35

[0190] As shown in the above results, when other generations were used as validation sets to verify the model, the four-point dataset model Model A accurately determined the relative cell generations of SIAISi011-A P10 and P20, indicating that the model-determined cell generations were close to the actual cell generations. Furthermore, the three-point dataset models Model B and Model C validation sets accurately determined cell generations within the modeling range, with Model B accurately determining the SIAISi011-A P10 generation and Model C accurately determining the SIAISi011-A P20 generation. Therefore, it can be inferred that as long as the re-input cell generation falls within the range of the pre-input cell generation, the model can accurately determine the re-input cell generation. In other words, as long as the pre-input cell generation range is sufficiently broad, the cell generation identification method of the present invention can accurately determine any cell generation from the same source.

[0191] In addition, the accuracy of the judgment results of Model A of the four-point data set was analyzed. The cell generations identified by the model were compared with the actual cell generations. The cell generations judged by the model were counted as correct if they were within ±2 generations of the actual cell generations, otherwise they were counted as incorrect. By comparing the two, the accuracy of the model was obtained as follows: Accuracy = Number of correctly judged cells / Total number of cells × 100%. The total accuracy of the model for the modeling set was 97.03%. The accuracy of the results verified by using other generations of cells from the same source reached 83.63%, and the total accuracy was 91.40%.

[0192] Similarly, to verify the universality of this method, the inventors used another source cell line, TUSMi002 (P5±2, P10±2, P15±2, P20±2, P25±2, P30±2, P35±2), to perform the same model establishment and verification. TUSMi002 P5, P15, P25, and P35 data were pre-input as a modeling dataset of known generations for comparison. Based on the number of pre-input datasets, two models were obtained: a four-point dataset (pre-input P5-P15-P25-P35, referred to as Model AT) and a three-point dataset (pre-input P5-P15-P25, referred to as Model BT, and pre-input P15-P25-P35, referred to as Model CT). The P5, P15, P25, and P35 data were re-input into the "cell generation determination model" in turn, combined with the similarity between the GBDT classification model and each group of data in the modeling dataset and the cell generation inference information given by the above-mentioned "feature extraction and comparison model." The final conclusion was that the model can accurately determine the cell generation of the input modeling group, and the model can be used as a TUSMi002 cell generation determination model (Figure 5, Table 5).

[0193] Table 5 Model validation results of the TUSMi002 model building group

[0194] Note: Model AT: TUSMi002 P5-15-25-35; Model BT: TUSMi002 P5-15-25; Model CT: TUSMi002 P15-25-35;

[0195] Other generations of TUSMi002 P10, P20, and P30 from the same source were used as the cell dataset to be identified to test the accuracy of Model AT, Model BT, and Model CT in determining cell generations. The results are shown in Figure 6 and Table 6.

[0196] Table 6. Model determination results of the TUSMi002 model to be identified

[0197] Note: Model AT: TUSMi002 P5-15-25-35; Model BT: TUSMi002 P5-15-25; Model CT: TUSMi002 P15-25-35;

[0198] The above results show that Model AT can accurately determine cell generations P10, P20, and P30, while Model BT accurately inferred cell generations P10 and P20, and Model CT accurately inferred cell generations P20 and P30. In other words, Model BT and Model CT can accurately determine cell generations within their respective modeling ranges. This result reaffirms the previous conclusion that, as long as the pre-input cell generation range is sufficiently broad, the "cell generation determination model" of the present invention can accurately determine any cell generation from the same source. Analysis and calculation of the accuracy of the "wide-range model" Model AT showed an overall accuracy of 97.88% for the modeling set. Verification results using cells of other generations from the same source achieved an accuracy of 64.60%, for a total accuracy of 83.57%.

[0199] In summary, a "cell generation identification model" composed of SR-FTIR spectral characteristic peak information was established and verified through the classification model algorithm. Given a cell line from any source, the model is established by collecting the SR-FTIR spectral characteristic peak information of cells of different generations within a sufficient range. This allows the relative physiological status of cells of other generations from the same source to be determined with high accuracy, achieving rapid, panoramic, and standardized quality evaluation.

[0200] Example 3

[0201] In this embodiment, the characteristic peaks involved in the model are deeply analyzed, and characteristic peak regions with high weights are selected to simplify the model.

[0202] The similarity of the characteristic peaks representing the information of lipid, protein, and nucleic acid regions was calculated by KL divergence, which is a method to measure the degree of match between two distributions. First, the characteristic peak data was normalized, and then the KL divergence calculation was performed on each pair of characteristic peaks. The similarity between the two was determined to be high using the standard of KL < 0.1. The minimum characteristic peak data set required for the "cell generation determination model" (hereinafter referred to as the minimum characteristic peak set) was screened to the greatest extent to obtain a simplified "cell generation determination model". The inventors found that the combination of the characteristic peaks α-helical of Amide I and NH bending of Amide I or α-helical of Amide I and β-pleated sheet of Amide I can highly characterize the cellular information of SIAISi011-A. At the same time, they found that the combination of the four characteristic peaks of νas(CH3), NH bending of Amide I, α-helical of Amide I, and α-helical of Amide II can highly characterize the cellular information of TUSMi002. Finally, we took the combined set of characteristic peaks screened from SIAISi011-A and TUSMi002 and obtained νas(CH3), NH bending of Amide I, α-helical of Amide I, and α-helical of Amide II, a minimum set of characteristic peaks that can best characterize cellular information. To confirm whether this minimum set of characteristic peaks could be used to model a simplified "cell generation determination model," we applied this set of characteristic peaks to determine cell generation based on the "cell generation determination model." Using the same method, we plotted the model-determined cell generation distribution histograms and model-determined cell generation and probability scatter plots (Figures 7 and 8) (Tables 7-10).

[0203] Table 7 Results of cell generation determination in the SIAISi011-A simplified model modeling group

[0204] Table 8 Results of cell generation determination in the TUSMi002 simplified model modeling group

[0205] Table 9 Results of determination of cell generation in the SIAISi011-A simplified model to be identified

[0206] Table 10 Results of cell generation determination in the TUSMi002 simplified model to be identified

[0207] The simplified "cell generation determination model" was established by applying the screened minimum characteristic peak set. The results of this simplified model for both the modeling group and the to-be-identified group were close to the actual cell generation, and no completely wrong conclusions were obtained, indicating that the simplified modeling method is feasible. Thereafter, according to the evaluation criteria described above, the determination results were evaluated and the accuracy of the model was calculated. The inventors found that the simplified "cell generation determination model" still had a high accuracy rate. The accuracy rate of the SIAISi011-A modeling group was 84.76%, the accuracy rate of the to-be-identified group was 77.78%, and the total accuracy rate was 81.82%; the accuracy rate of the TUSMi002 modeling group was 92.37%, the accuracy rate of the to-be-identified group was 56.75%, and the total accuracy rate was 77.05%.

[0208] This result proves that the combination of the four characteristic peaks of νas(CH3), NH bending of Amide I, α-helical of Amide I, and α-helical of Amide II is the minimum characteristic peak set that can be used to establish a "cell generation determination model", further supporting that the "minimum characteristic peak set" of single-cell SR-FTIR obtained by this model can be used for the subsequent invention of a spectrometer for rapid SR-FTIR acquisition. At the same time, the determined and evaluated physiological state of the cell can provide a basis for clinical cell treatment.

[0209] Example 4

[0210] In this example, cross-validation was used to verify the ability of the cell generation identification model system to identify cell generations from different sources.

[0211] The inventors have determined that, given a cell line from any source, a cell generation identification model can accurately identify any cell generation within the modeling range. Furthermore, they have developed "large-scale models"—Model A and Model AT—based on cells from two different sources (SIAISi011-A and TUSMi002), each of which can accurately determine the generation of cells from other sources. However, the ability of these cell generation identification models to identify the generation of cells from other sources has not yet been validated. In other words, for the identification of the generation of cells from any source, is it necessary to establish a "large-scale model" of its own source, or can it be validated using known cell models from other sources, such as Model A or Model AT?

[0212] To address this issue, the inventors first cross-validated the cell data from the Model A and Model AT modeling groups. Model A was used to identify the relative cell generations of TUSMi002 P5, P15, P25, and P35, and Model AT was used to identify the relative cell generations of SIAISi011-A P5, P15, P25, and P35. The results are shown in Figure 9, Table 11, and Table 12.

[0213] Table 11: Model A identification results of TUSMi002 modeling group cell generations

[0214] Table 12: Model AT identification results of SIAISi011-A modeling group cell generations

[0215] The large-scale model Model A established by SIAISi011-A cannot accurately identify the cell generation of the TUSMi002 modeling group. Similarly, the large-scale model Model AT established by TUSMi002 cannot accurately identify the cell generation of the SIAISi011-A modeling group. Therefore, it is believed that the cell generation identification model cannot accurately identify the cell generation of other sources.

[0216] To verify this result again, the inventors used Model A and Model AT to cross-verify the cell data outside the modeling group, that is, Model A was used to identify the relative cell generations of TUSMi002 P10, P20, and P30, and Model AT was used to identify the relative cell generations of SIAISi011-A P10, P20, and P30. The results are shown in Figure 10, Table 13, and Table 14.

[0217] Table 13: Model A identification results of TUSMi002 test group cell generations

[0218] Table 14: Model AT identification results of SIAISi011-A test group cell generations

[0219] It was also found that Model A was difficult to accurately identify the cell generations of TUSMi002 P10, P20, and P30. At the same time, Model AT was also unable to identify the cell generations of the SIAISi011-A test group.

[0220] In summary, this cell generation identification model can accurately identify any cell generation within the modeling range of its own source, but it cannot identify cell generations from other sources. The conclusion is that to identify cell generations from any source, it is necessary to collect SR-FTIR spectra from a wide range of cell generations, select characteristic peaks, and apply the cell generation identification model to establish a "wide-range model."

[0221] Example 5

[0222] In this embodiment, the XGBoost algorithm was used to establish a cell generation identification model.

[0223] While the biological significance of spectral absorption peaks at many specific vibrational frequencies remains undefined, several have been identified as well-characterized indicators of intracellular material content and structural composition (hereafter collectively referred to as characteristic peaks). To achieve unbiased, standardized extraction of characteristic peak information, we developed the YY algorithm. This algorithm first identifies a sufficient wavenumber range around the occurrence of a target characteristic peak. Within this range, a model is constructed using the XGBoost algorithm based on metrics such as the overall number of peaks, the peak order of the target characteristic peak, the peak wavenumber, and peak morphology. This standardized extraction of characteristic peak information within a specific frequency band is achieved using XGBoost. XGBoost, a gradient boosting tree algorithm, boasts high prediction accuracy and can automatically handle nonlinear and complex relationships between features, leading to a better data fit. XGBoost also handles missing values ​​and outliers and exhibits robustness to noise. Using these features, we established a model for each characteristic peak extraction. Comparing this model with manually extracted characteristic peak information, we achieved an overall accuracy exceeding 95%, achieving full-band extraction of biologically meaningful characteristic peak information.

[0224] Subsequently, we applied the SR-FTIR spectral peak data set to develop a machine learning model capable of determining cell passage number. We used the previously collected second-order derivative spectral peak data for each passage of SIAISi011-A and TUSMi002-A as the training set for model training. We also collected SR-FTIR spectra at P10, P20, and P30 of continuous culture and derived 27 sets of second-order derivative spectral peak information from these spectra as the test set.

[0225] XGBoost is also used to train the model. The specific process is that we obtain the average, median, variance and other information indicators of each feature peak of the modeling group through random sampling, and apply XGBoost for model training.

[0226] We then fed the validation set data into the model. The generation of a single cell was determined by randomly sampling the cell with another random cell. The trained model was then used to determine the predictions for each sample. The median of the predicted results was taken as the predicted outcome, and the confidence interval between the two generations above and below was used as the confidence probability. This resulted in the single-cell generation determination result and the probability of the result. The modeling flow chart is shown in Figure 11.

[0227] Table 15 shows the 27 characteristic peaks screened by this method.

[0228] Table 15

[0229] Those skilled in the art will appreciate that the above-mentioned embodiments are specific examples for implementing the present invention, and that in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A method for identifying the generation of a cell population comprising stem cells, It is characterized in that The method comprises the steps of: Determine the probability that the cell population to be identified is in each of the N known generations based on a preset generation identification model; Comparing the characteristic values ​​of the cell population to be identified with the characteristic values ​​of N cell populations of known generations, respectively, to obtain the generation range to which the cell population to be identified belongs; The generation of the cell population to be identified is obtained according to the range of generations to which the cell population to be identified belongs and the probability that the cell population to be identified is in each generation.

2. The method according to claim 1, It is characterized in that The preset generation identification model is constructed by the following method: Constructing a training data set, the training data set comprising characteristic quantities of N cell populations with known passage numbers and corresponding passage number values, wherein the characteristic quantities of the cell populations comprise several wavelength values ​​obtained by SR-FTIR analysis of the cell populations; Constructing an initial identification model, wherein the initial identification model includes an independent variable and a dependent variable, wherein the independent variable is a characteristic value of the cell population, and the dependent variable is a probability of the cell population being in each generation; The initial identification model is trained based on the training data set to obtain a final generation identification model.

3. The method according to claim 2, It is characterized in that The characteristic quantity of the cell group includes νas(CH 3 ), NH bending of Amide I, α-helical of Amide I and α-helical of Amide II and the corresponding displacement values.

4. The method according to claim 1, It is characterized in that The characteristic quantities of the cell population to be identified and the characteristic quantities of the N known generations of cell populations are subjected to dimensionality reduction processing, and then the two are compared to obtain the generation range to which the cell population to be identified belongs.

5. The method according to claim 4, It is characterized in that Performing cluster analysis on the characteristic quantities of the cell population to be identified after the dimension reduction process to obtain non-outlier points of the cell population to be identified; and Performing cluster analysis on the characteristic quantities of the N known generations of cell populations that have undergone dimensionality reduction processing, respectively, to obtain cluster centers of the N known generations of cell populations; By comparing the distances between the non-outlier points of the cell population to be identified and the cluster centers of the N cell populations of known generations, the generation range to which the cell population to be identified belongs is obtained.

6. The method according to claim 5, It is characterized in that By comparing the distances between the non-outlier points of the cell population to be identified and the cluster centers of the N cell populations of known generations, at least two target cluster centers are selected from the cluster centers of the N cell populations of known generations, and the generation range is formed according to the generation of the cell population of known generations to which the target cluster centers belong.

7. The method according to claim 5, It is characterized in that Selecting a first target cluster center that is closest to the non-outlier point of the cell population to be identified and a second target cluster center that is second closest to the non-outlier point of the cell population to be identified from the cluster centers of the N known generations of the cell population; and forming the generation range using the generations of the cell population of known generations to which the first target cluster center and the second target cluster center belong; and / or, selecting a first target cluster center which is closest to the non-outlier point of the cell population to be identified from among the cluster centers of the N known generations of cell populations; Selecting cluster centers of cells of two known generations adjacent to the generation to which the first target cluster center belongs, comparing the distances between the cluster centers of the two adjacent cells of known generations and the non-outlier points of the cell population to be identified, and selecting the cluster center of cells of the known generation that is closer to the non-outlier points of the cell population to be identified as the second target cluster center; The generation range is composed of the generation of the cell population of known generation to which the first target cluster center and the second target cluster center belong.

8. The method according to claim 1, It is characterized in that The generation range of the cell population to be identified and the probability that the cell population to be identified is in each generation are calculated using a weighted average formula to obtain the generation of the cell population to be identified.

9. A marker panel for identifying the generation of a cell population, It is characterized in that The marker group includes the following characteristic peaks: νas(CH 3 ), NH bending of Amide I, α-helical of Amide I and α-helical of Amide II.

10. A method for identifying the generation of a cell population, It is characterized in that The method comprises the steps of: A cell to be identified is selected from the cell population to be identified as a target cell, and N cells are randomly selected from the cell population to be identified to form a first cell cluster with the target cell. Obtaining characteristic quantities of cells in the first cell cluster, and using a preset generation identification model to make a determination to obtain a first identification generation value corresponding to the target cell; Randomly extract N cells from the identified cell group again to form a second cell group with the target cells. Obtaining the characteristic value of the second cell cluster, and using a preset generation identification model to make a determination, to obtain a second identification generation value corresponding to the target cell; By repeating this cycle, a plurality of identification generation values ​​corresponding to the target cells are obtained, and data processing is performed on the plurality of identification generation values ​​to obtain the generation of the cell population to be identified.

Citation Information

Patent Citations

  • Identification of polymers and dispersants in rubber adhesive R80

    CN105548061A

  • Gene detection kit for identifying passage number of bone marrow mesenchymal stem cells based on circular RNA

    CN108192959A

  • Long noncoding RNA based gene detection kit for identifying passage number of umbilical cord mesenchymal stem cell

    CN108441548A

  • Method and marker for identifying generations of stem cell population

    CN116312813A

  • Automated Classification of Cells in Biologic Mixtures Analyzed by High Parameter Cytometry Instrumentation, Processing, System and Method

    US20160125134A1