Data processing method and system for tumor cell detection

Abstract

The invention discloses a data processing method and system for tumor cell detection, and relates to the technical field of data processing.The method comprises the steps that cells are sequenced, and cell metabolism is monitored; performing time delay reconstruction according to the embedding dimension and the delay time, mapping gene expression data to a reconstruction phase space, and calculating a chaos degree of gene expression; establishing a biomarker white list, inputting the chaos degree into an adaptive weight distributor, setting feature retention weights for markers in the white list, constructing a protective auto-encoder, dynamically adjusting the dimension of a bottleneck layer, and performing forced constraint when the markers in the white list are detected; the method comprises the following steps: extracting coupling features through metabolic oscillation data and physical microenvironment data, calculating gradient sensitivity of shear stress to metabolic oscillation, carrying out feature fusion on the coupling features and gene expression features, and constructing a subtype classification boundary for the fused features through feature vector selection in combination with the gradient sensitivity. Accurate dimension reduction of data and deep analysis of biological characteristics are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a data processing method and system for tumor cell detection. Background Art

[0002] In biomedical research, especially in the fields of oncology and cell biology, the processing and analysis of high-dimensional data plays a vital role. However, in practical applications, due to the high dimensionality of the data, dimensionality reduction or feature compression is often required for more efficient data storage, transmission and analysis. However, this process may be accompanied by a significant loss of key biological information, bringing challenges to biological analysis.

[0003] Traditional dimensionality reduction methods, such as principal component analysis (PCA) and linear discriminant analysis (LDA), can reduce data dimensions to a certain extent, but may ignore nonlinear relationships and dynamic characteristics in the data, resulting in the loss of key biological information. For example, in tumor research, tumor stem cell markers such as CD133 are of great significance for understanding the occurrence, development and treatment of tumors. However, in the process of dimensionality reduction, if the method is inappropriate, the expression characteristics of these key markers may be annihilated, thereby weakening the biological analysis ability of the data.

[0004] As a commonly used feature compression method, autoencoders learn low-dimensional representations of data by training a neural network. However, when autoencoders are over-compressed, they may also lead to the loss of key biological information. In addition, during the compression process, autoencoders may not be able to effectively distinguish between cell subtypes with completely different biological characteristics, such as invasive subtypes and inert subtypes. These subtypes may be incorrectly mixed and displayed in low-dimensional space, resulting in blurred or even overlapping boundaries between subtypes, which greatly troubles subsequent biological analysis and understanding. Summary of the invention

[0005] 1. Technical issues to be resolved In view of the shortcomings of the prior art, the present invention provides a data processing method and system for tumor cell detection. Single cells are captured by a microfluidic chip, and gene expression, metabolic oscillation and physical microenvironment data are combined. The pseudo nearest neighbor method is used to determine the embedding dimension, and a protective autoencoder is constructed to retain key biomarker information. Finally, the features are fused to construct subtype classification boundaries, thereby achieving accurate dimensionality reduction of the data and in-depth analysis of biological characteristics.

[0006] (II) Technical solution To achieve the above objectives, the present invention is implemented through the following technical solutions: a data processing method for tumor cell detection, comprising: Sequence cells and monitor cell metabolism to obtain gene expression and metabolic oscillation data, quantify the physical microenvironment of cells, and obtain physical microenvironment data; The embedding dimension is determined by the pseudo nearest neighbor method, the delay time is determined by the first zero crossing of the autocorrelation function, the time delay is reconstructed according to the embedding dimension and the delay time, the gene expression data is mapped to the reconstructed phase space, and the chaos degree of gene expression is calculated; Establish a whitelist of biomarkers, input the chaos degree into the adaptive weight allocator, set feature retention weights for the markers in the whitelist, build a protective autoencoder, dynamically adjust the bottleneck layer dimension, input gene expression data into the protective autoencoder, extract gene expression features, and enforce constraints when whitelist markers are detected; Coupling features were extracted from metabolic oscillation data and physical microenvironment data, and the gradient sensitivity of shear stress to metabolic oscillation was calculated. The coupling features and gene expression features were fused, and the subtype classification boundaries were constructed for the fused features through feature vector selection combined with gradient sensitivity.

[0007] Furthermore, physical microenvironment data is obtained, including: scanning the cell surface, obtaining the three-dimensional stiffness matrix data of the cell, and calculating the shear stress of the cell in the flow field based on particle image velocimetry in the microfluidic chip: ,in, represents the cell surface shear stress, m Indicates the viscosity of the medium, v Indicates speed, x Represents the position along the flow direction.

[0008] Furthermore, the gene expression data is mapped to the reconstructed phase space, and the reconstruction formula is: ,in, F(t) represents the reconstructed gene expression data, G (t) represents gene expression data, t Indicates the time, Indicates the delay time. m represents the embedding dimension; In the reconstructed phase space, calculate the distance evolution model of each pair of adjacent points: , where d represents the distance and t represents the time. represents the initial distance, l represents the slope; Take the logarithm of the distance evolution model: , after linearly fitting the slope of the distance evolution model of adjacent point pairs, the average of the slopes of all adjacent point pairs is taken as the chaos degree.

[0009] Furthermore, the embedding dimension is determined by the pseudo nearest neighbor method, including: Obtain gene expression data; select a smaller initial embedding dimension m; calculate the nearest neighbor of each data point under the current embedding dimension m; increase the embedding dimension m by 1, construct a new high-dimensional space, and map the gene expression data to the high-dimensional space; in the newly constructed high-dimensional space, calculate the nearest neighbor of each data point, if the nearest neighbor changes after the dimension is increased, then the nearest neighbor is judged to be a false neighbor, and the proportion of false neighbors is calculated; if the rate of change of the proportion of false neighbors exceeds a preset proportion threshold, continue to increase the embedding dimension m until the rate of change of the proportion of false neighbors does not exceed the preset proportion threshold; when the rate of change of the proportion of false neighbors does not exceed the preset proportion threshold, the corresponding embedding dimension m is selected as the final embedding dimension.

[0010] Further, the delay time is determined by the first zero crossing of the autocorrelation function, including: For gene expression data, calculate its autocorrelation function: ,in, R(t) represents the autocorrelation function, Indicates the delay time. G(t) represents gene expression data, n Indicates the number of data points; Find the first zero crossing point on the autocorrelation function and correspond the first zero crossing point to value as the delay time.

[0011] Furthermore, an adaptive weight allocator is constructed, the chaos degree is input into the adaptive weight allocator, and the feature retention threshold is calculated: , where T represents the feature retention threshold, σ represents the sigmoid function, represents the degree of chaos, represents the baseline chaos degree, Indicates the initial threshold; sets the feature retention weight for the markers in the whitelist: ,in, oh represents the feature preservation weight.

[0012] Furthermore, a protective autoencoder is constructed, the bottleneck layer dimension is set as an adaptive variable, the feature retention weight and chaos degree are input into the bottleneck layer, and the dynamic compression rate is calculated: , where DCR represents the dynamic compression ratio; calculate the bottleneck layer dimension: ,in, Represents the bottleneck layer dimension value, Represents the baseline dimension; Gene expression data is fed into the protective autoencoder, and when whitelist markers are detected, the constraints are enforced .

[0013] Furthermore, the metabolic oscillation data and the physical microenvironment data are subjected to tensor product operations to extract coupling features: ,in, Indicates i The metabolic oscillation value at each time point, is the three-dimensional stiffness matrix, j and k denote the rows and columns of the three-dimensional stiffness matrix, respectively. Represents coupling characteristics; Calculate the gradient sensitivity of shear stress to metabolic oscillations: ,in, c represents the gradient sensitivity, represents the cell surface shear stress, v Indicates speed, x represents the position along the flow direction, M represents the metabolic oscillation value, and t represents the time.

[0014] Furthermore, the coupling features and gene expression features are fused, and the subtype classification boundary is constructed for the fused features through feature vector selection: , where X represents the feature vector selection, c represents the gradient sensitivity, represents the maximum shear stress difference, Represents metabolic response delay time.

[0015] A data processing system for tumor cell detection, comprising: The data acquisition module sequences cells and monitors cell metabolism to obtain gene expression and metabolic oscillation data, quantifies the physical microenvironment of cells, and obtains physical microenvironment data; The gene expression analysis module determines the embedding dimension through the pseudo nearest neighbor method, determines the delay time through the first zero crossing of the autocorrelation function, reconstructs the time delay according to the embedding dimension and the delay time, maps the gene expression data to the reconstructed phase space, and calculates the chaos degree of gene expression; The autoencoder module builds a whitelist of biomarkers, feeds the chaos degree into an adaptive weight allocator, sets feature retention weights for markers in the whitelist, builds a protective autoencoder, dynamically adjusts the bottleneck layer dimension, feeds gene expression data into the protective autoencoder, extracts gene expression features, and enforces constraints when whitelist markers are detected; The feature selection module extracts coupling features through metabolic oscillation data and physical microenvironment data, calculates the gradient sensitivity of shear stress to metabolic oscillation, fuses the coupling features and gene expression features, combines gradient sensitivity, and constructs subtype classification boundaries for the fused features through feature vector selection.

[0016] (III) Beneficial effects The present invention provides a data processing method and system for tumor cell detection, which has the following beneficial effects: (1) By acquiring high-resolution gene expression data, real-time monitoring of cell metabolism, and quantification of the cell's physical microenvironment, we can have a more comprehensive understanding of the biological characteristics of tumor cells, including information on their gene expression, metabolic state, and physical microenvironment. This helps to reduce the loss of key biological information during dimensionality reduction or feature compression, and improve the accuracy and depth of biological analysis.

[0017] (2) Through the pseudo nearest neighbor method and the first zero crossing of the autocorrelation function, the embedding dimension and delay time can be accurately determined, and then the gene expression data can be accurately mapped to the reconstructed phase space, thereby more realistically reflecting the dynamic behavior and potential nonlinear characteristics of the data. The calculation of the chaos degree helps to understand the complexity and instability of gene expression.

[0018] (3) The feature retention weights are dynamically adjusted through an adaptive weight allocator, and then the bottleneck layer dimension of the autoencoder is dynamically adjusted. The dimensionality reduction strategy can be flexibly adjusted according to the complexity of the data and the distribution of key information, ensuring that key biological markers such as tumor stem cells are given priority retention, avoiding the loss of key information that may be caused during dimensionality reduction or feature compression. The gene expression environment data is input into the protective autoencoder. When whitelist markers are detected, the mandatory constraints further enhance the accuracy and reliability of data processing.

[0019] (4) By extracting metabolic-physical coupling features and calculating the gradient sensitivity of shear stress to metabolic oscillations, a deeper understanding of the biological characteristics of tumor cells was achieved; feature fusion and feature vector selection further improved the accuracy of subtype classification, which helped to accurately distinguish key subtypes such as invasive subtypes and indolent subtypes, providing strong support for subsequent biological analysis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a schematic diagram of the steps of the data processing method for tumor cell detection of the present invention; Figure 2 Schematic diagram of the autoencoder network structure of the present invention; Figure 3 The figure is a schematic diagram of the structure of the data processing system for tumor cell detection according to the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] See also Figure 1-Figure 2 The present invention provides a data processing method for tumor cell detection, comprising the following steps: Step 1: Sequence the cells and monitor cell metabolism to obtain gene expression and metabolic oscillation data, quantify the cell physical microenvironment, and obtain physical microenvironment data; The step 1 includes the following contents: Step 101: Use a microfluidic chip (channel width of 20 μm) to capture single cells. The microfluidic chip precisely controls the flow of fluid so that cells pass through narrow channels one by one and are captured. The capture process must ensure the survival rate of the cells and avoid cell damage during the capture process; Step 102: Setting the lysis solution pulse trigger frequency, such as triggering a lysis solution pulse (lasting 5 ms) every 30 seconds, to lyse the captured single cells, and using SMART-seq2 technology to generate full-length cDNA, ensuring that the sequencing depth of each cell reaches or exceeds 50,000 reads / cell, and the time resolution reaches the minute level, thereby providing high-resolution gene expression data; Step 103: Integrate a fluorescence resonance energy transfer (FRET) biosensor (such as iGlucose) to monitor the intracellular glucose uptake rate in real time, and set the sampling frequency, such as 100 Hz, to capture fast metabolic oscillation data. The collected data is subjected to noise reduction processing through Kalman filtering to improve the accuracy of the data; Step 104: Scan the cell surface using an atomic force microscope (AFM, probe stiffness is 0.1 N / m) to obtain the three-dimensional stiffness matrix data of the cell, and calculate the shear stress on the cell in the flow field based on the particle image velocimetry (PIV) technology in the microfluidic chip: ,in, represents the cell surface shear stress, m Indicates the viscosity of the medium, v Indicates speed, x Indicates the position along the flow direction; Step 105: Generate an encrypted timestamp to ensure that the gene expression, metabolism and physical data are aligned in time to microsecond accuracy, perform eigenvalue analysis on the three-dimensional stiffness matrix, the eigenvalue analysis includes but is not limited to the definition method, characteristic polynomial method or iteration method, if the eigenvalue is greater than the characteristic threshold, such as 15kPa, then trigger the weighted filtering processing of the gene expression data, the weighted filtering processing includes but is not limited to linear weighted filtering, exponential weighted filtering or weighted median filtering; the characteristic threshold is the basis for judging whether the weighted filtering processing needs to be triggered, and the characteristic threshold is determined according to the cell type, experimental conditions and data analysis requirements.

[0023] When using, combine the contents of step 101 to step 105: By acquiring high-resolution gene expression data, real-time monitoring of cell metabolism, and quantification of the cell's physical microenvironment, we can more comprehensively understand the biological characteristics of tumor cells, including information on their gene expression, metabolic state, and physical microenvironment. This helps reduce the loss of key biological information during dimensionality reduction or feature compression, and improves the accuracy and depth of biological analysis.

[0024] Step 2: Determine the embedding dimension by the pseudo nearest neighbor method, determine the delay time by the first zero crossing of the autocorrelation function, perform time delay reconstruction based on the embedding dimension and delay time, map the gene expression data to the reconstructed phase space, and calculate the chaos degree of gene expression; The step 2 includes the following contents: Step 201: Determine the embedding dimension by the pseudo nearest neighbor method, determine the delay time by the first zero crossing of the autocorrelation function, perform time delay reconstruction according to the embedding dimension and the delay time, and map the gene expression data to the reconstructed phase space. The reconstruction formula is: ,in, F(t) represents the reconstructed gene expression data, G (t) represents gene expression data, t Indicates the time, Indicates the delay time. m represents the embedding dimension; Among them, the embedding dimension is determined by the pseudo nearest neighbor method, including: Data preparation: Obtain gene expression data; Initial embedding dimension selection: Choose a small initial embedding dimension m, usually starting from 2 or 3; Calculate the nearest neighbor points: Under the current embedding dimension m, calculate the nearest neighbor points of each data point by calculating the Euclidean distance or other distance metrics; Construct a high-dimensional space: increase the embedding dimension m by 1, construct a new high-dimensional space, and map the original data into the high-dimensional space; Calculate the proportion of false neighbor points: In the newly constructed high-dimensional space, calculate the nearest neighbor point for each data point. If the nearest neighbor point changes after the dimension increases, that is, it is not the same as the nearest neighbor point in the previous dimension, then the nearest neighbor point is judged as a false neighbor point, and the proportion of false neighbor points is calculated; Determine whether the embedding dimension is sufficient: If the rate of change of the proportion of false neighbor points exceeds the preset ratio threshold, it means that the current embedding dimension m may not be sufficient to fully capture the dynamic behavior of the data. Therefore, it is necessary to continue to increase the embedding dimension m and repeat the above steps. If the rate of change of the proportion of false neighbor points does not exceed the preset ratio threshold, then the current embedding dimension m can be considered sufficient; In the pseudo nearest neighbor method, the rate of change of the proportion of false neighbor points refers to the change of the proportion of false neighbor points relative to the previous embedding dimension as the embedding dimension increases. This rate of change is used to determine whether the current embedding dimension is large enough to fully capture the dynamic behavior in the data. The ratio threshold is calculated by gradually increasing the embedding dimension and calculating the rate of change of the proportion of false neighbor points. The trend of the rate of change as the embedding dimension increases is observed. According to the trend and the initially estimated threshold range, the threshold is gradually adjusted until a suitable value is found. Determine the final embedding dimension: When the change rate of the proportion of false neighbor points does not exceed the preset proportion threshold, the corresponding embedding dimension m is selected as the final embedding dimension; The delay time is determined by the first zero crossing of the autocorrelation function, including: For gene expression data, calculate its autocorrelation function: ,in, R(t) represents the autocorrelation function, Indicates the delay time. G(t) represents gene expression data, n Indicates the number of data points; On the graph of the autocorrelation function, find the position of the first zero crossing and correspond the position of the first zero crossing to value as the delay time; Step 202: In the reconstructed phase space, calculate the distance evolution model of each pair of adjacent points: , where d represents the distance and t represents the time. represents the initial distance, l represents the slope; Step 203: Taking the logarithm of the distance evolution model: , after linearly fitting the slope of the distance evolution model of adjacent point pairs, the average of the slopes of all adjacent point pairs is taken as the chaos degree ; The degree of chaos is calculated by calculating the slope of the distance evolution model of adjacent point pairs (i.e. l), which provides an overall measure of the chaotic behavior in the entire data set. It should be noted that the degree of chaos is a relative concept, which reflects the sensitivity of the system to the initial conditions. When used, combine the contents of step 201 to step 203: Through the pseudo nearest neighbor method and the first zero crossing method of the autocorrelation function, the embedding dimension and delay time can be accurately determined, and then the gene expression data can be accurately mapped to the reconstructed phase space, thereby more realistically reflecting the dynamic behavior and potential nonlinear characteristics of the data. The calculation of chaos degree helps to understand the complexity and instability of gene expression.

[0025] Step 3: Establish a whitelist of biomarkers, input the chaos degree into the adaptive weight allocator, set feature retention weights for the markers in the whitelist, build a protective autoencoder, dynamically adjust the bottleneck layer dimension, input gene expression data into the protective autoencoder, extract gene expression features, and enforce constraints when whitelist markers are detected; The step three includes the following contents: Step 301: Establish a biomarker whitelist database (including tumor stem cell markers such as CD133), use a bidirectional long short-term memory (LSTM) network to extract gene expression features, set the number of memory units to 128, and the time window sliding step to 5 minutes; Step 302: Construct an adaptive weight allocator, input the chaos degree into the adaptive weight allocator, and calculate the feature retention threshold: , where T represents the feature retention threshold, σ represents the sigmoid function, represents the degree of chaos, represents the baseline chaos degree, Represents the initial threshold. The baseline chaos degree is a preset value used to standardize the chaos degree. The initial threshold is a parameter adjusted according to experimental data and requirements. Set feature retention weights for markers in the whitelist: ,in, oh represents the feature retention weight; Step 303: Construct a protective autoencoder network, set the bottleneck layer dimension as an adaptive variable, input the feature retention weight and chaos degree into the bottleneck layer control module, and calculate the dynamic compression rate: , where DCR represents the dynamic compression ratio; calculate the bottleneck layer dimension: ,in, Represents the bottleneck layer dimension value, Indicates the base dimension, which is a preset reference value of the dimension (such as 32 dimensions); Step 304: Feed the gene expression data into the protective autoencoder, and enforce the constraint when the whitelist marker is detected ; When used, combine the contents of step 301 to step 304: By dynamically adjusting the feature retention weights through an adaptive weight allocator, and then dynamically adjusting the bottleneck layer dimension of the autoencoder, the dimensionality reduction strategy can be flexibly adjusted according to the complexity of the data and the distribution of key information, ensuring that key biological markers such as tumor stem cells are given priority retention, avoiding the loss of key information that may occur during dimensionality reduction or feature compression, and inputting gene expression environment data into the protective autoencoder. When whitelist markers are detected, the mandatory constraints further enhance the accuracy and reliability of data processing.

[0026] Step 4: Extract coupling features through metabolic oscillation data and physical microenvironment data, calculate the gradient sensitivity of shear stress to metabolic oscillation, fuse the coupling features and gene expression features, combine gradient sensitivity, and construct subtype classification boundaries for the fused features through feature vector selection.

[0027] The step 4 includes the following contents: Step 401: Establishing a metabolic-physical coupling tensor: Performing a tensor product operation on the metabolic oscillation data and the physical microenvironment data to extract metabolic-physical coupling features: ,in, Indicates i The metabolic oscillation value at each time point, is the three-dimensional stiffness matrix, j and k denote the rows and columns of the three-dimensional stiffness matrix, respectively. Represents coupling characteristics; Step 402: Calculate the gradient sensitivity of shear stress to metabolic oscillations: ,in, c represents the gradient sensitivity, represents the cell surface shear stress, v Indicates speed, x represents the position along the flow direction, M represents the metabolic oscillation value, and t represents the time; Step 403: The metabolic-physical coupling features and the gene expression features are fused by the splicing method, and the subtype classification boundary is constructed for the fused features by feature vector selection: , where X represents the feature vector selection, c represents the gradient sensitivity, represents the maximum shear stress difference (per unit length along the flow direction), represents the metabolic response delay time (the time lag from shear stress change to metabolic oscillation response); When used, combine the contents of step 401 to step 402: By extracting metabolic-physical coupling features and calculating the gradient sensitivity of shear stress to metabolic oscillations, a deeper understanding of the biological characteristics of tumor cells was achieved; feature fusion and feature vector selection further improved the accuracy of subtype classification, helping to accurately distinguish key subtypes such as invasive subtypes and inert subtypes, providing strong support for subsequent biological analysis and treatment.

[0028] See also Figure 3 The present invention also provides a data processing system for tumor cell detection, comprising: a data acquisition module, a gene expression analysis module, an autoencoding module and a feature selection module; wherein, The data acquisition module sequences cells and monitors cell metabolism to obtain gene expression and metabolic oscillation data, quantifies the physical microenvironment of cells, and obtains physical microenvironment data; The gene expression analysis module determines the embedding dimension through the pseudo nearest neighbor method, determines the delay time through the first zero crossing of the autocorrelation function, reconstructs the time delay according to the embedding dimension and the delay time, maps the gene expression data to the reconstructed phase space, and calculates the chaos degree of gene expression; The autoencoder module builds a whitelist of biomarkers, feeds the chaos degree into an adaptive weight allocator, sets feature retention weights for markers in the whitelist, builds a protective autoencoder, dynamically adjusts the bottleneck layer dimension, feeds gene expression data into the protective autoencoder, extracts gene expression features, and enforces constraints when whitelist markers are detected; The feature selection module extracts coupling features through metabolic oscillation data and physical microenvironment data, calculates the gradient sensitivity of shear stress to metabolic oscillation, fuses the coupling features and gene expression features, combines gradient sensitivity, and constructs subtype classification boundaries for the fused features through feature vector selection.

[0029] In the application, the several formulas involved are all calculated by removing dimensions and taking their numerical values. The formula is a formula of the most recent real situation obtained by collecting a large amount of data and performing software simulation. The coefficients in the formula are set by technicians in this field according to actual conditions.

[0030] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in combination with electronic hardware, computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0031] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0032] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A data processing method for tumor cell detection, characterized in that: include: Sequence cells and monitor cell metabolism to obtain gene expression and metabolic oscillation data, quantify the physical microenvironment of cells, and obtain physical microenvironment data; The embedding dimension is determined by the pseudo nearest neighbor method, the delay time is determined by the first zero crossing of the autocorrelation function, the time delay is reconstructed according to the embedding dimension and the delay time, the gene expression data is mapped to the reconstructed phase space, and the chaos degree of gene expression is calculated; Establish a whitelist of biomarkers, input the chaos degree into the adaptive weight allocator, set feature retention weights for the markers in the whitelist, build a protective autoencoder, dynamically adjust the bottleneck layer dimension, input gene expression data into the protective autoencoder, extract gene expression features, and enforce constraints when whitelist markers are detected; Coupling features were extracted from metabolic oscillation data and physical microenvironment data, and the gradient sensitivity of shear stress to metabolic oscillation was calculated. The coupling features and gene expression features were fused, and the subtype classification boundaries were constructed for the fused features through feature vector selection combined with gradient sensitivity.

2. The data processing method for tumor cell detection according to claim 1, characterized in that: Obtain physical microenvironment data, including: scanning the cell surface, obtaining the three-dimensional stiffness matrix data of the cell, and calculating the shear stress of the cell in the flow field based on particle image velocimetry in the microfluidic chip: ,in, represents the cell surface shear stress, μ Indicates the viscosity of the medium, v Indicates speed, x Indicates the position along the flow direction.

3. The data processing method for tumor cell detection according to claim 1, characterized in that: Map gene expression data to the reconstruction phase space, reconstruction formula: ,in, F(t) represents the reconstructed gene expression data, G(t) represents gene expression data, t Indicates the time, Indicates the delay time. m represents the embedding dimension; In the reconstructed phase space, calculate the distance evolution model of each pair of adjacent points: ,in, d Indicates distance, represents the initial distance, λ represents the slope; Take the logarithm of the distance evolution model: , linearly fit the slope of the distance evolution model of adjacent point pairs, and take the average of the slopes of all adjacent point pairs as the chaos degree.

4. The data processing method for tumor cell detection according to claim 3, characterized in that: The embedding dimension is determined by the pseudo nearest neighbor method, including: Obtain gene expression data; select an initial embedding dimension m; calculate the nearest neighbor of each data point under the current embedding dimension m; increase the embedding dimension m by 1, construct a new high-dimensional space, and map the gene expression data to the high-dimensional space; in the new high-dimensional space, calculate the nearest neighbor of each data point. If the nearest neighbor changes after the dimension is increased, the nearest neighbor is judged to be a false neighbor, and the proportion of false neighbors is calculated; if the rate of change of the proportion of false neighbors exceeds a preset proportion threshold, continue to increase the embedding dimension m until the rate of change of the proportion of false neighbors does not exceed the preset proportion threshold; when the rate of change of the proportion of false neighbors does not exceed the preset proportion threshold, the corresponding embedding dimension m is selected as the final embedding dimension.

5. The data processing method for tumor cell detection according to claim 4, characterized in that: The delay time is determined by the first zero crossing of the autocorrelation function, including: For gene expression data, calculate its autocorrelation function: ,in, R(τ) represents the autocorrelation function, n Indicates the number of data points; Find the first zero crossing point on the autocorrelation function and correspond the first zero crossing point to τ value as the delay time.

6. The data processing method for tumor cell detection according to claim 1, characterized in that: Construct an adaptive weight allocator, input the chaos degree into the adaptive weight allocator, and calculate the feature retention threshold: ,in, T represents the feature retention threshold, σ represents the sigmoid function, represents the degree of chaos, represents the baseline chaos degree, Indicates the initial threshold; sets the feature retention weight for the markers in the whitelist: ,in, ω represents the feature preservation weight.

7. The data processing method for tumor cell detection according to claim 6, characterized in that: Construct a protective autoencoder, input the feature retention weight and chaos degree into the bottleneck layer, and calculate the dynamic compression rate: ,in, DCR Indicates the dynamic compression rate; sets the bottleneck layer dimension as an adaptive variable and calculates the bottleneck layer dimension: ,in, Represents the bottleneck layer dimension value, Represents the baseline dimension; Gene expression data is fed into the protective autoencoder, and when whitelist markers are detected, the constraints are enforced .

8. The data processing method for tumor cell detection according to claim 1, characterized in that: Perform tensor product operations on metabolic oscillation data and physical microenvironment data to extract coupling features: ,in, Indicates i The metabolic oscillation value at each time point, is the three-dimensional stiffness matrix, j and k denote the rows and columns of the three-dimensional stiffness matrix, respectively. Represents coupling characteristics; Calculate the gradient sensitivity of shear stress to metabolic oscillations: ,in, γ represents the gradient sensitivity, M represents the metabolic oscillation value, t Indicates time.

9. The data processing method for tumor cell detection according to claim 8, characterized in that: The coupling features and gene expression features are fused, and the subtype classification boundary is constructed for the fused features through feature vector selection: ,in, X represents feature vector selection, γ represents the gradient sensitivity, represents the maximum shear stress difference, Represents metabolic response delay time.

10. A data processing system for tumor cell detection, used to implement the method according to any one of claims 1 to 9, characterized in that: include: The data acquisition module sequences cells and monitors cell metabolism to obtain gene expression and metabolic oscillation data, quantifies the physical microenvironment of cells, and obtains physical microenvironment data; The gene expression analysis module determines the embedding dimension through the pseudo nearest neighbor method, determines the delay time through the first zero crossing of the autocorrelation function, reconstructs the time delay according to the embedding dimension and the delay time, maps the gene expression data to the reconstructed phase space, and calculates the chaos degree of gene expression; The autoencoder module builds a whitelist of biomarkers, feeds the chaos degree into an adaptive weight allocator, sets feature retention weights for markers in the whitelist, builds a protective autoencoder, dynamically adjusts the bottleneck layer dimension, feeds gene expression data into the protective autoencoder, extracts gene expression features, and enforces constraints when whitelist markers are detected; The feature selection module extracts coupling features through metabolic oscillation data and physical microenvironment data, calculates the gradient sensitivity of shear stress to metabolic oscillation, fuses the coupling features and gene expression features, combines gradient sensitivity, and constructs subtype classification boundaries for the fused features through feature vector selection.

Images