A method for determining cellular order parameters at the transcriptome phase transition boundary

By using high-throughput transcriptome sequencing or gene chip detection, the mean expression level and the square of the coefficient of variation of genes are calculated. A bivariate distribution piecewise linear regression is used to locate the stability wall, which solves the problem that existing technologies cannot objectively divide the cell order intervals and realizes cell state assessment across species and conditions.

CN123120028APending Publication Date: 2026-11-17HANGZHOU AOKANG YOUAN BIOTECHNOLOGY CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611101050.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-23
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies lack quantifiable phase transition boundaries when assessing cell state, making it impossible to objectively and automatically divide the core and peripheral regions of the genome. This results in the inability to make absolute comparisons of cell order indicators across different experiments and species, and the inability to identify dynamic phase transitions and critical windows during the order establishment process.

Method used

By performing high-throughput transcriptome sequencing or gene chip detection on the samples, the mean expression level and the squared coefficient of variation of each gene are calculated. Piecewise linear regression is performed using a bivariate distribution to automatically locate the stability wall, divide the core region and the outer region, and calculate the absolute cellular order parameters through the order core gene set.

Benefits of technology

It achieves cross-species and cross-condition universality, can accurately identify the critical window of order establishment or collapse, provides absolutely comparable cellular order parameters, and improves the accuracy and information integrity of cellular state assessment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120028A_ABST
    Figure CN123120028A_ABST
Patent Text Reader

Abstract

This invention proposes a method for determining cellular order parameters at the transcriptome phase transition boundary, belonging to the field of parameter measurement technology. By performing high-throughput transcriptome sequencing or gene chip detection on samples, the mean and squared coefficient of variation of the expression level of each gene in multiple samples or replicates are calculated. Piecewise linear regression or sliding window fitting is performed on the genes using bivariate distribution. The core and peripheral regions of the genes are automatically divided using stability wall information and fitting residuals. The proportion of genes in the core region or the first principal component is calculated using the core gene set of the order system for samples or time points to obtain absolute cellular order parameters. This invention enables the determination of absolute cellular order parameters across species and experimental conditions, and can automatically identify dynamic phase transition critical windows during cell differentiation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention proposes a method for determining cellular order parameters at the transcriptome phase transition boundary, which relates to the field of parameter determination technology, specifically to the field of cellular order parameter determination technology based on the transcriptome noise-expression level phase transition boundary. Background Technology

[0002] Accurate assessment of cell state is a cornerstone of the biomedical and biopharmaceutical fields, widely used in scenarios such as stem cell differentiation determination, recombinant protein production monitoring, and tumor progression assessment. Existing technologies mainly include principal component analysis, average expression levels of marker genes, gene set enrichment fractions, and direct measurement of single-cell entropy or noise. These existing technologies are limited in scope, relying solely on average gene expression levels while completely ignoring the crucial physical dimension of expression noise, leading to severe distortions in the characterization of cell state. They lack quantifiable phase transition boundaries, making it impossible to objectively and automatically delineate the core and peripheral regions of the genome, thus hindering absolute comparisons of cell order indicators across different experiments and species. They fail to identify dynamic phase transitions and critical windows during the establishment of order, missing optimal opportunities for prediction and intervention; and they lack cross-species and cross-condition universality. Summary of the Invention

[0003] This invention provides a method for determining cellular order parameters at the transcriptome phase transition boundary to solve the above-mentioned problems: This invention proposes a method for determining cellular order parameters at the transcriptome phase transition boundary, the method comprising: S1. By performing high-throughput transcriptome sequencing or gene chip analysis on the samples, the expression levels of each gene in multiple samples or replicates are calculated using the mean and coefficient of variation squared to obtain the log-mean and log-CV of each gene. 2 a bivariate distribution; S2. By performing piecewise linear regression or sliding window fitting on the genes through bivariate distribution, the phase transition boundary, i.e., the stability wall information and its confidence interval, is obtained. S3. Automatically divide the genes into core and peripheral regions using stable wall information and fitting residuals to obtain an ordered core gene set; S4. Calculate the percentage of genes in the core region or the first principal component of the sample or time point using the core gene set of the order to obtain the absolute cellular order parameters.

[0004] Furthermore, the system includes: The variable distribution module is used to calculate the mean and coefficient of variation (CV) of the expression level of each gene in multiple samples or replicates by performing high-throughput transcriptome sequencing or gene chip detection on the samples, thereby obtaining the log-mean and log-CV of each gene. 2 a bivariate distribution; The phase transition boundary module is used to obtain the phase transition boundary, i.e., the stability wall information and its confidence interval, by performing piecewise linear regression or sliding window fitting on genes through bivariate distribution. The automatic partitioning module is used to automatically partition genes into core and peripheral regions using stable wall information and fitting residuals to obtain an ordered core gene set. The parameter acquisition module is used to calculate the proportion of genes in the core region or the first principal component of a sample or time point using the core gene set of the order, thereby obtaining absolute cellular order parameters.

[0005] The beneficial effects of this invention are as follows: Existing technologies (such as PCA, marker gene average expression method, gene set enrichment analysis) only utilize the mean or rank of gene expression levels, completely discarding the intercellular coefficient of variation (CV) of gene expression. 2 This invention, for the first time, uses a double logarithmic scatter plot to analyze the order information contained within the expression level, treating it as a dimension equivalent to the expression level. This innovation allows the order parameter to simultaneously reflect two fundamental physical quantities: expression intensity (whether it is sufficiently active) and expression stability (whether there is coordination and consistency among cells). Experiments have shown that cells in a high-order state (such as terminally differentiated β cells and high-producing CHO cells) not only have high expression levels of genes in their core regions, but also exhibit high intercellular coefficients (CVs). 2 Significantly reduced; while low-order cells (such as stem cells and stress cells) exhibit high CV. 2 Current technology cannot capture this silent order.

[0006] Existing technologies for distinguishing core and peripheral genes either rely on subjectively set expression thresholds (e.g., TPM>1) or use quantiles (e.g., the top 10%) that lack physical meaning. These thresholds are not comparable across different datasets and laboratories and lack error estimation. This invention uses piecewise linear regression to automatically locate the point of most drastic derivative change as a stability wall. This point is mathematically unique and can be provided with a 95% confidence interval via Bootstrap. The biological interpretation of the stability wall is clear: it is the critical expression level at which a gene transitions from a high-noise / low-expression state to a low-noise / high-expression state. This boundary consistently appears across different species and sequencing platforms (yeast: log). 10 (TPM)≈-0.20 (i.e., TPM≈0.63), people: log 10 (TPM)≈-1.16 (i.e., TPM≈0.069), Arabidopsis thaliana: log 10 (TPM)≈-1.67 (i.e. TPM≈0.021)), which proves its universal physical regularity.

[0007] Existing order indices (such as PC1 and enrichment score) are relative values ​​and cannot be directly compared between different experiments. The order parameter α(t) provided by this invention has the following characteristics that allow for absolute comparison: Proportional method: α = number of genes in the core region / total number of genes, ranging from 0 to 1, directly quantifying the proportion of genes in an ordered state. This value can be directly compared between different samples and different species. For example, α≈0.25 when human pancreatic β cells differentiate and mature, α≈0.18 in the oxidative phase of the yeast metabolic cycle, and α≈0.32 in high-yielding CHO cell lines.

[0008] PCA method: Although PC1 is a relative value, by fixing the gene set of the core region (based on stable wall screening), the direction and scale of PC1 remain consistent across different datasets, avoiding the directional drift caused by changes in the gene set in traditional PCA.

[0009] After extensive verification, this invention is applicable to: Species: yeast (Saccharomyces cerevisiae), mammals (human, mouse, and Chinese hamster ovary CHO cells), and plants (Arabidopsis thaliana) all successfully localized stable walls.

[0010] Sequencing platforms: Illumina bulk RNA-seq, 10x single-cell RNA-seq, and Affymetrix microarray. For single-cell data, a stable wall was also obtained using the sliding window method.

[0011] Tissue types: cell lines (HUVEC, CHO), primary tissues (human islets, mouse embryonic pancreas), in vitro differentiation systems (hPSC→β cells).

[0012] Sample size: Only 3 biological replicates are required to calculate CV. 2 When the number of repetitions is less than 3, an external reference wall or sliding window can be used instead.

[0013] This universality means that the invention can be used as a standardized tool, eliminating the need for retraining or calibration for each new system, thus greatly reducing the application threshold.

[0014] The kurtosis of the stable wall defined in this invention (α_sharp = -k2) is extremely sensitive to the cellular energy metabolism state. In the HUVEC serum starvation experiment: Normal culture (0h): kurtosis ≈ 11.6, wall located at x0 ≈ -1.12 (log 10 CPM); 24-hour starvation: Steepness ≈ 9.8, wall moved to the right to x0 ≈ -1.21 (log 10 CPM); 72 hours of starvation: Steepness only 2.7, the wall moves further to the right; 24 hours after recovery: Steepness increased to 9.6, and the wall shifted to the left.

[0015] This indicates that steepness can serve as an index of cell health or metabolic activity. Note: The absolute value of this slope (κ=-k2) is a different physical quantity from the order parameter α (the proportion of genes in the core region) mentioned later, and should not be confused. In contrast, the PCA score or marker gene expression levels in existing technologies change laggingly and insignificantly during this process. The dynamic wall tracking capability of this invention also allows for the automatic identification of critical windows for the establishment or collapse of order (such as the S phase in cell differentiation, the inflection point of stress tolerance). Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a method for determining cellular order parameters at the transcriptome phase transition boundary. Detailed Implementation

[0017] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0018] In one embodiment of the present invention, a method for determining cellular order parameters at transcriptomic phase transition boundaries is provided, the method comprising: S1. By performing high-throughput transcriptome sequencing or gene chip detection on the samples, the expression level of each gene in multiple samples or replicates is calculated using the mean and coefficient of variation squared to obtain the bivariate distribution of the log mean and log CV² of each gene. S2. By performing piecewise linear regression or sliding window fitting on the genes through bivariate distribution, the phase transition boundary, i.e., the stability wall information and its confidence interval, is obtained. S3. Automatically divide the genes into core and peripheral regions using stable wall information and fitting residuals to obtain an ordered core gene set; S4. By calculating the proportion of genes in the core region or the first principal component of the sample or time point using the core gene set, the absolute cellular order parameters are obtained, such as... Figure 1 As shown.

[0019] Collect cell or tissue samples for testing, extract total RNA, and obtain raw expression data using high-throughput sequencing (RNA-seq, including single-cell RNA-seq and bulk RNA-seq) or gene chip technology. For each sample, record its biological background information (such as species, cell type, treatment conditions, time point, and biological repeat number).

[0020] The sequencing reads are compared and quantified to obtain the expression level of each gene. One of the following units can be used: TPM (Transcripts Per Million): Eliminates the influence of gene length and sequencing depth; FPKM (FragmentsPerKilobaseoftranscriptperMillionmappedreads) or RPKM; Raw counts are suitable for subsequent models based on the negative binomial distribution; For chip data, the fluorescence intensity signal is used directly.

[0021] If multiple technical or biological replicates exist, batch effect correction (such as ComBat, Harmony, etc.) must be performed first. However, if the variability among replicate samples is mainly contributed by biological noise, correction may not be necessary.

[0022] To ensure statistical robustness, genes with expression levels of zero (or below the detection threshold) in more than 90% of the samples are filtered out. For single-cell data, it is recommended that each gene be expressed in at least three cells.

[0023] For each gene g, there are n samples (or replicates) with an observed expression level of e. g ,1,e g ,2,...,e g , n ,but: Gene mean:

[0024] Standard deviation:

[0025] Coefficient of variation squared:

[0026] Among them, e g This represents the arithmetic mean of the expression level of gene g. The standard deviation of gene g expression level. This represents the squared coefficient of variation of gene g, i.e., transcriptional noise.

[0027] To avoid division by zero, when At that time, place The gene was then removed from subsequent analyses.

[0028] For mean and CV 2 Perform a base-10 logarithmic transformation and add a small constant ε (usually 10⁻⁶) to avoid zero or negative infinity:

[0029] Where, x g y represents the logarithmic mean expression of gene g. gε represents the squared logarithmic coefficient of variation of gene g, and ε is an added minimal constant used to avoid taking the logarithm for zero or negative infinity, usually taken as 10⁻⁶.

[0030] All genes (x) g ,y g (x) g ,y g The points are plotted in a Cartesian coordinate system, forming a bivariate distribution of the logarithmic mean and the logarithmic coefficient of variation squared.

[0031] For chip data or standardized expression values, ε can be increased appropriately (e.g., 10⁻³).

[0032] All gene points (x) g y g Plotted in a Cartesian coordinate system. It is typically observed that as x(log0) increases... 10 As the mean increases, y(log) 10 CV 2 The expression level initially decreases gradually or slowly, then drops sharply after a certain point. This inflection point is the stability wall, marking the critical expression level at which a gene moves from the outer region (low expression / high noise) to the core region (high expression / low noise).

[0033] The data is fitted using two linear functions, and continuity is forced at the breakpoint x0.

[0034] Where x0 represents the location of the stability wall, i.e. the critical expression threshold for the gene to enter the core region from the outer region; y0 represents the expected logarithmic coefficient of variation squared value at x=x0; k1 represents the slope of the low expression region on the left, which is usually close to zero or positive (the noise increases slightly or remains flat as the expression level increases); k2 represents the slope of the high expression region on the right, which is constrained to be negative (the higher the expression level, the faster the noise decreases).

[0035] The parameters {x0, y0, k1, k2} are solved using nonlinear least squares methods (such as Levenberg-Marquardt) or global optimization algorithms (such as differential evolution and simulated annealing), minimizing the sum of squared residuals.

[0036] Constraints: k2 < 0, and k1 ≥ -0.5 can be selected (to avoid excessive sloping to the left). Initial values ​​can be set as: x0 = median(x), y0 = median(y), k1 = 0, k2 = -1.

[0037] To assess the robustness of the stable wall location, the Bootstrap method is used: Resample the gene set with replacement to generate B=500 simulated datasets (each dataset has the same number of genes as the original set).

[0038] Repeat the above steps for each simulated dataset to obtain... .

[0039] Calculate the 95% confidence interval for x0: ,in Let q be the quantile of the Bootstrap distribution.

[0040] The sharpness of a stable wall is defined as the absolute value of the slope on the right side: αsharp = -k² (the larger the positive value, the steeper the wall). Steepness can reflect the urgency of system order: steepness usually increases during differentiation or stress recovery.

[0041] For each gene, calculate its actual y. g Compared with piecewise linear model predictions The difference:

[0042] in From the x of this gene g It is calculated based on the fitted x0, y0, k1, k2.

[0043] A gene that simultaneously meets both of the following conditions is defined as a core region: Expression level higher than the stability wall: x g >x0 Negative residual: r g <0 (meaning the actual noise is lower than the noise predicted by the model) All other genes (including X) g ≤x0 or x g Genes with >x0 but positive residuals are defined as periphery.

[0044] To improve the reliability of genes in the core region, a significance test can be performed on the negativeness of the residuals: Null hypothesis: The residuals do not significantly deviate from zero.

[0045] Method: A permutation test was used—the y-values ​​were randomly shuffled 1000 times, and the model was refitted and the residuals calculated each time to obtain a zero-distribution residual. If the true residual r... g If the value is less than the 5th percentile of the distribution, then the significance is considered to be p<0.05.

[0046] Only genes with negative residuals satisfying p<0.05 are retained as the core region.

[0047] Genes in the core region were subjected to GO (GeneOntology) or KEGG enrichment analysis to verify whether they were enriched in basic functional categories such as ribosomes, translation, and oxidative phosphorylation; genes in the peripheral region should be enriched in categories such as signal transduction, development, and stress response, thereby confirming the biological rationality of the partitioning.

[0048] For each sample to be tested (such as a bulkRNA-seq sample at a certain time point), extract the expression vector of the core region genes in the sample (length = number of core region genes m).

[0049] The order parameter α can be calculated using any of the following methods: Method A (Proportion Method): α=m / M Where M is the total number of genes after filtering. α ranges from 0 to 1, representing the proportion of genes located in the ordered core. This method is suitable for cases where the total number of genes is consistent across different samples and is directly comparable across different datasets.

[0050] Method B (PCA method): Principal component analysis was performed on the expression matrix (sample × gene) of the core region genes, and the first principal component (PC1) was taken as α. α = PC1. If PC1 is contrary to the expected biological direction (e.g., PC1 should increase during differentiation but decreases), it is multiplied by -1. This method can comprehensively reflect the synergistic expression intensity of genes in the core region.

[0051] Method C (Average Expression):

[0052] Where E j This is the standardized expression level of gene j (e.g., logTPM). This method is simple, but it is easily dominated by a few highly expressed genes.

[0053] For a series of samples collected along the time axis or by processing gradients, calculate α(t) for each sample to obtain the curve of α(t) as a function of t. If α(t) undergoes a drastic change within a certain time window (e.g., jumps from 0.1 to 0.8), then that window corresponds to the phase transition interval of the system order.

[0054] When using single-cell RNA-seq data, each cell is typically considered a sample. However, the expression matrix of a single cell is too sparse, making direct calculation of the CV (expression coefficient) impractical. 2 We need to utilize group information. A sliding window method can be used: For single-cell RNA-seq data, cells are sorted by pseudo-time or real time, and w consecutive cells are taken as a window (w=50). Within the window, the mean and coefficient of variation squared for each gene are calculated:

[0055] Using the time / position of the window center point as the x-axis, a bivariate point set for each window is obtained, and then the stability wall location and order parameters are calculated according to methods three to seven.

[0056] Where w represents the number of cells contained in the window. This represents the mean expression level of gene g within the window. This represents the squared coefficient of variation of gene g within the window.

[0057] Sort the cells by pseudo-time or real time, and take w consecutive cells (e.g., w=50) as a window.

[0058] Calculate the mean and CV for each gene within the window. 2 (Cells within the window are considered duplicates).

[0059] Using the time / position of the window's center point as t, we obtain the set of x and y points for each window.

[0060] The process for stabilizing the wall and defining the core area is the same as that for bulk data.

[0061] For α(t) values ​​from different experimental batches, if direct comparison is required, the following normalization strategy is recommended: Using α as a reference (e.g., the maximum proportion of β cells in adult pancreatic islets), scale α to the range of 0 to 1.

[0062] Alternatively, a relative order parameter can be used: αrel=(α-αbaseline) / (αmax-αbaseline), where αbaseline is the α of the early undifferentiated state.

[0063] The calculated α(t) was correlated with known biological phenotypes (such as cell differentiation rate, antibody yield, survival rate, and biomarker expression level) using a correlation analysis (Pearson or Spearman) to verify the accuracy of α as a cell state indicator.

[0064] For data containing multiple time points, steps 2 and 3 can be repeated to obtain the stable wall position x0(t) and kurtosis αsharp(t) at each time point. Observe the changes in these parameters as development or disease progresses: Moving the wall to the right indicates that a higher level of expression is needed to enter the core area, which is a sign of improved order.

[0065] Increased steepness indicates a more rapid decrease in noise in high-expression regions, and a tendency for the cell population to become more homogeneous.

[0066] The method of this invention was applied to datasets from different species (yeast, mice, humans, plants, etc.) to compare the distribution range of stable wall locations. If the walls were found to be located at almost the same expression level (e.g., log10(TPM)≈0.5~0.8), it demonstrates that the method has cross-species universality.

[0067] This method utilizes the yeast (Saccharomyces cerevisiae) metabolic cycle dataset from the publicly available database GSE52339. This dataset contains 16 consecutive time points, sampled every 25 minutes, covering a complete metabolic cycle of approximately 400 minutes (about 1.5 cycles). Each time point has 3 biological replicates, totaling 48 samples. The sequencing platform is Illumina HiSeq, and raw reads are normalized to TPM (Transcripts Per Million).

[0068] Download the standardized gene expression matrix (genes × sample) from the GEO database. Genes with TPM < 0.1 in all samples were filtered out (considered as unexpressed), leaving approximately 5,200 genes for analysis. For the three replicates at each time point, the following steps were performed: Extract the three repeated expression values ​​at this time point.

[0069] For each gene, calculate the arithmetic mean (mean) and standard deviation (SD) of expression levels in three replicates.

[0070] Calculate the squared coefficient of variation: CV 2 =(SD / Mean) 2 When Mean=0, CV 2 Set as missing.

[0071] Perform a logarithmic transformation: x = log 10 (Mean+1e-6), y=log 10 (CV 2 +1e-6), where 1e-6 is a small constant to avoid zero values.

[0072] Data was collected at the 100-minute time point (corresponding to the early reduction phase of the metabolic cycle). The (x, y) points of approximately 5,200 genes were plotted on a two-dimensional coordinate system. Observations show that when x < -0.5, the y values ​​are distributed between -1.5 and 0.5; when x > -0.2, the y values ​​drop sharply to below -1.5. A piecewise linear regression model was used.

[0073] The parameters were solved using the Levenberg-Marquardt algorithm, with initial values ​​set as follows: x0 = median(x) = -0.3, y0 = median(y) = -0.8, k1 = 0, k2 = -1.2. The constraint k2 < 0 was applied. Fitting results: X0 = -0.20 (95% confidence interval obtained through 500 Bootstrap resampling: [-0.44, -0.02]) Y0=-0.67 K1=0.12 (slight increase on the left, indicating a slight increase in noise in the low expression region) K2 = -1.85 (a steep drop on the right indicates that noise in the high-expression region decays rapidly) Steepness α_sharp=-k2=1.85.

[0074] Calculate the predicted value for each gene based on the fitted piecewise straight line. and residual Genes that meet the following two conditions will be marked as core regions: x>x0 (expression level is higher than the stability wall, i.e., log) 10 (TPM) > -0.20, corresponding to original TPM > 0.63) r < 0 (actual noise is lower than model prediction) Ultimately, 511 core region genes were identified (approximately 9.8% of the total analyzed genes). Functional enrichment analysis (DAVID) showed that core region genes were significantly enriched in basal metabolic processes such as ribosomal proteins (e.g., RPS1A, RPL4B), translation elongation factors (TEF1, TEF2), and glycolytic enzymes (ENO2, PGK1) (p<1e-20, FDR corrected). Peripheral region genes were enriched in functions such as signal transduction, stress response, and transcriptional regulation.

[0075] The above-described stabilizing wall localization and core region division were repeated for all 16 time points. However, to maintain consistency of the order parameters, the 511 core region genes selected at the 100-minute time point were consistently used as the reference gene set (because core region genes are relatively conserved throughout the metabolic cycle). Then: For each time point, the expression matrix of these 511 core region genes was extracted (the average of 3 replicates, i.e., a 1×511 vector for each time point).

[0076] Principal component analysis (PCA) was performed on the expression matrices (16×511) at all 16 time points. The first principal component (PC1) was taken as the order parameter α(t). Since PC1 showed a high value in the oxidative phase (OX), consistent with the known metabolically active state, no sign inversion was performed.

[0077] Plot the curve of α(t) as a function of time. The results show: In the oxidized phase (OX, approximately 0–50 min, 200–250 min), α(t) reaches its peak value (approximately 1.5–2.0). In the reduced phase RB (approximately 75–125 min), α(t) drops to a trough (approximately -1.2). α(t) rises again in the reduced phase RC (approximately 175–225 min).

[0078] The oscillation period of α(t) is approximately 200 minutes, which is highly consistent with the known yeast metabolic cycle (approximately 200–250 minutes). Pearson correlation analysis of α(t) with the respiratory quotient (RQ) reported in the original literature showed a correlation coefficient r = 0.92 (p < 0.001), demonstrating that the order parameter measured in this invention can accurately reflect changes in cellular metabolic state.

[0079] This method validates the feasibility of the invention in a periodic oscillation system of a single-celled eukaryote (yeast), successfully distinguishing different phases in the metabolic cycle and quantitatively characterizing the dynamic changes in system order (α(t)). The biological functions of the core region genes are consistent with expectations, demonstrating that the stability wall can correctly identify the highly expressed, low-noise order center.

[0080] This method utilizes the human umbilical vein endothelial cell (HUVEC) dataset from the publicly available database GSE213323. Experimental design: Normally cultured HUVEC cells (0h), serum starvation for 24h and 72h, and serum recovery culture for 24h after 72h starvation (R24h). Three biological replicates were used for each condition, for a total of 12 samples. The sequencing platform was Illumina NextSeq500, and FPKM values ​​were output.

[0081] Download the FPKM expression matrix and filter out genes with FPKM < 0.1 in all samples, leaving approximately 14,000 genes. Process each condition (0h, 24h, 72h, R24h) separately: Take 3 duplicate samples under this condition.

[0082] Calculate the mean, standard deviation, and CV for each gene. 2 and log 10 (Mean+1e-6) and log 10 (CV2 +1e-6) transformation.

[0083] For each time condition, the complete process of stabilizing wall positioning in Example 1 was independently repeated (piecewise linear regression + Bootstrap 500 times). The parameters for each condition were obtained: condition <![CDATA[wall position x0(log 10 FPKM)]]> 95% CI <![CDATA[Steepness α_sharp=-k2]]> 0h (normal) 0.18 [0.12,0.24] 11.6 24-hour hunger 0.31 [0.25,0.38] 5.8 48-hour starvation 0.39 [0.32,0.47] 3.9 72 hours of starvation 0.46 [0.39,0.54] 2.7 Restore 24h 0.25 [0.19,0.31] 9.6 It was observed that with prolonged serum starvation time, the stable wall position shifted to the right from -1.12 (0h) through -1.21 (24h) to 3.41 (72h), and the kurtosis significantly decreased from 11.6 to 2.7. After 24h of serum recovery culture, the wall position shifted back to -1.13, and the kurtosis recovered to 9.6 (but was still lower than 11.6 in the normal group, suggesting that some damage was irreversible).

[0084] This can be revised to: It was observed that as serum starvation time increased, the stable wall position x0 gradually shifted to the right from 0.18 (normal) to 0.46 (72h starvation), and the kurtosis significantly decreased from 11.6 to 2.7. After 24h of serum recovery, x0 shifted back to 0.25, and the kurtosis increased back to 9.6. To examine the changes in cell order under different treatment conditions, the proportional method was used to calculate α(t): Use the core region gene set for each condition (instead of a fixed reference set) because starvation can alter the core region composition.

[0085] α = number of genes in the core region / total number of genes analyzed (14,000).

[0086] result: 0h: α=0.078 (approximately 1,092 core genes) 24-hour starvation: α = 0.051 (approximately 714 individuals) 72-hour starvation: α = 0.027 (approximately 378 individuals) Recovery time 24 hours: α = 0.062 (approximately 868) The α value continued to decrease with prolonged starvation, and partially rebounded after recovery, consistent with the trend of wall position movement and steepness changes.

[0087] GO enrichment analysis of the core region genes in the normal group (0h) showed enrichment in pathways such as angiogenesis, cell adhesion, and oxidative phosphorylation. After 72h of starvation, the number of core region genes decreased, and the enriched pathways changed to endoplasmic reticulum stress, ubiquitination degradation, and apoptosis, reflecting the transition of cells from homeostasis to stress and apoptosis. The core region genes in the recovery group partially reverted to the characteristics of the normal group, but a small number of stress genes remained.

[0088] This method demonstrates that the present invention can dynamically track the evolution of transcriptional order in mammalian cells during nutritional stress and recovery. The location, steepness, and order parameter α of the stability wall are highly sensitive to energy state and possess biological interpretability. This method can serve as a quantitative tool for assessing cellular health and stress response, for example, for drug toxicity screening or optimization of tissue preservation conditions.

[0089] The working principle and technical effects of the above-mentioned technical solution are as follows: This method obtains raw gene expression data by collecting cell or tissue samples, extracting total ribonucleic acid, and performing high-throughput sequencing or gene chip detection. The arithmetic mean and standard deviation of the expression level of each gene in multiple samples or replicates are calculated, and then the coefficient of variation squared is calculated to quantify the degree of expression fluctuation (i.e., transcriptional noise) of each gene. After performing a base-10 logarithmic transformation on the mean expression level and the coefficient of variation squared, and adding a minimal constant, the logarithmic mean and logarithmic coefficient of variation squared for each gene are obtained, thus constructing a bivariate distribution for all genes. In this distribution, as the logarithmic mean increases, the logarithmic coefficient of variation squared shows a trend of first leveling off and then decreasing sharply. This invention automatically locates this inflection point of decrease using a piecewise linear regression model and defines it as the stability wall. Based on the location of the stability wall and the fitting residuals, genes with expression levels higher than the stability wall and actual noise lower than the model prediction are divided into the core region, and the remaining genes are divided into the peripheral region. By calculating the proportion of genes in the core region or performing principal component analysis on the gene expression matrix in the core region, absolute cellular order parameters that can be directly compared across samples and species can be obtained.

[0090] This method addresses the distortion in cell state characterization caused by existing techniques that rely solely on mean expression levels while completely ignoring the noise dimension. It also overcomes the technical deficiency of lacking objective, quantifiable phase transition boundaries to distinguish the core and periphery regions. Compared to existing techniques, this method significantly improves the accuracy and completeness of cell order assessment, transforming previously ignored noise dimensions into quantifiable order indicators. By introducing mathematically definable stability walls, it achieves a leap from subjective threshold-based classification to objective, data-driven classification, greatly reducing the bias and uncertainty introduced by manually set thresholds. The resulting absolute order parameters enable comparisons of cell states across different experiments, species, and sequencing platforms, effectively enhancing the method's universality and comparability.

[0091] In one embodiment of the present invention, S1 includes: By collecting cell or tissue samples to be tested and extracting total ribonucleic acid, the samples are subjected to high-throughput sequencing or gene chip detection to obtain raw expression data; By comparison and quantification, the original expression level of each gene in multiple samples or replicates is quantified and standardized using the number of transcripts per million, the number of mapped readings per kilobase per million, or the original counting units to obtain the observed expression level. The expression mean and standard deviation are obtained by calculating the arithmetic mean and standard deviation of the expression levels of each gene in all samples or replicates. The coefficient of variation squared for each gene is obtained by dividing the standard deviation of each gene by the mean expression level and then squaring the result. By performing a base-10 logarithmic transformation on the mean expression level and the squared coefficient of variation of each gene, and adding a very small constant, the logarithmic mean and the logarithmic coefficient of variation of each gene are obtained. By combining the logarithmic mean and the squared logarithmic coefficient of variation of all genes, a bivariate distribution of the logarithmic mean and the squared logarithmic coefficient of variation is obtained.

[0092] The working principle and technical effects of the above-mentioned technical solution are as follows: This method collects samples and extracts total ribonucleic acid, and obtains raw expression data through high-throughput sequencing or gene chip detection. Through comparison and quantification, the expression level of each gene in multiple samples or replicates is standardized using the number of transcripts per million, the number of mapped readings per kilobase per million, or the raw counting units, to obtain observed expression levels. The arithmetic mean and standard deviation of the observed values ​​for each gene in all samples or replicates are calculated to obtain the expression mean and standard deviation. The standard deviation is divided by the mean and squared to obtain the coefficient of variation squared, which quantifies the degree of expression fluctuation of the gene among different samples. Subsequently, the mean and the coefficient of variation squared are logarithmically transformed to base 10 and a minimal constant is added to eliminate dimensional differences and avoid zero-value problems. Finally, the logarithmic mean and logarithmic coefficient of variation squared of all genes are combined to form a complete bivariate distribution.

[0093] This method addresses the incomparability of cross-sample and cross-batch data caused by the lack of standardized processing in existing technologies, as well as the bias resulting from directly using raw counts and ignoring the influence of gene length and sequencing depth. By applying various standardization methods to expression levels (number of transcripts per million, number of mapped reads per kilobase per million, etc.), it effectively eliminates the interference of differences in gene length and sequencing depth on expression level comparisons, improving the comparability of expression levels between different samples. By calculating the squared coefficient of variation, it quantifies gene expression noise for the first time into an independent dimension that can be analyzed equivalently to the expression level mean, filling the gap in existing technologies that ignore noise information. Logarithmic transformation and the addition of small constants make highly skewed expression data tend towards a normal distribution, significantly improving the stability and reliability of model fitting.

[0094] In one embodiment of the present invention, obtaining a bivariate distribution of the logarithmic mean and the logarithmic coefficient of variation by combining the logarithmic mean and the logarithmic coefficient of variation squared of all genes includes: By combining the logarithmic mean and the squared logarithmic coefficient of variation of all genes in a one-to-one correspondence according to the gene index, a two-dimensional coordinate point corresponding to each gene is obtained. By pooling the two-dimensional coordinates of all genes into a single dataset, a bivariate distribution of the log mean and the log-squared coefficient of variation is obtained.

[0095] Specifically, by combining the logarithmic mean and the squared logarithmic coefficient of variation of all genes in a one-to-one correspondence according to the gene index, a two-dimensional coordinate point corresponding to each gene is obtained, including: By using the log mean value and the log coefficient of variation squared value of each gene, we obtain the two log values ​​for each gene. The x-coordinate of each gene is obtained by taking the log mean of each gene as the x-coordinate value of that gene in the Cartesian coordinate system. The ordinate of each gene is obtained by taking the squared logarithmic coefficient of variation of each gene as the ordinate value of the corresponding gene in the Cartesian coordinate system. By pairing the x-coordinate value and y-coordinate value of each gene according to the gene's index or identifier, an ordered pair corresponding to each gene is obtained. By defining the ordered pair as a two-dimensional coordinate point of the gene in a bivariate distribution, a two-dimensional coordinate point is obtained for each gene. The coordinate point reflects both the expression intensity (x-axis) and expression stability (y-axis) of the gene.

[0096] The working principle and technical effect of the above technical solution are as follows: This method uses the obtained log-mean and log-variance coefficient squared values ​​of each gene. The log-mean of each gene is used as the x-coordinate value in a Cartesian coordinate system, and the log-variance coefficient squared is used as the y-coordinate value. According to the index or identifier of each gene, the x-coordinate values ​​and y-coordinate values ​​are paired one-to-one, forming an ordered pair for each gene. This ordered pair is defined as the two-dimensional coordinate point of the gene in the bivariate distribution, so that each gene is uniquely represented in two-dimensional space by both its expression intensity (x-coordinate) and expression stability (y-coordinate). By aggregating the two-dimensional coordinate points of all genes into the same dataset, a complete bivariate distribution is obtained.

[0097] This method addresses the limitation of existing technologies in simultaneously visualizing or analyzing the relationship between two different physical dimensions (expression intensity and expression stability). By mapping each gene to a coordinate point in a two-dimensional space, it achieves a holistic visualization of the expression level-noise relationship across the entire transcriptome, making phase transition phenomena, previously hidden in high-dimensional data, readily observable. The x-axis of each gene reflects its expression activity, while the y-axis reflects its expression stability; the combination of these two dimensions allows for precise location of each gene within the order hierarchy. This bivariate coordinate system construction provides a clear data organization structure for subsequent identification of stability walls and the division of core and peripheral regions, fundamentally solving the signal loss problem caused by single-dimensional analysis and significantly improving the information utilization efficiency of cellular order assessment.

[0098] In one embodiment of the present invention, S2 includes: A bivariate scatter plot is obtained by plotting the logarithmic mean and the squared logarithmic coefficient of variation of all genes in a Cartesian coordinate system. The bivariate distribution is fitted using a two-segment continuous piecewise linear regression model through a bivariate scatter plot, with the slope of the right segment constrained to be negative, thus obtaining the piecewise linear regression model. The location of the stable wall is obtained by solving the parameters in the piecewise linear regression model using nonlinear least squares method or global optimization algorithm; By resampling the gene set with replacement 500 times and repeating the above fitting process each time, the values ​​of 500 stable wall positions are obtained. By calculating the 0.025th and 0.975th quantiles of 500 stable wall locations, a 95% confidence interval for the stable wall locations is obtained. Steepness is obtained by calculating the absolute value of the slope on the right side; Complete information about the stable wall can be obtained by combining the location, confidence interval, and steepness of the stable wall.

[0099] The sliding window method is implemented as follows: Step 1: Sorting and Windowing First, sort all genes in ascending order of their logarithmic mean. Then, use a small window of fixed width (0.3 logarithmic units) to slide across the entire data range from left to right. At each position, retrieve the gene points contained within the window.

[0100] Step 2: Find the local median For each gene point retrieved from a window, we only look at the values ​​in the "logarithmic coefficient of variation squared" column and take the median of these values ​​as the representative value for that window position. In this way, the entire data range is simplified into a "median curve" formed by connecting the representative values ​​of each window.

[0101] Step 3: Further smooth the curve The median curve is not smooth enough. We use a smoothing filter (similar to using a short window for local polynomial fitting) to remove the small jitters on the curve, making it smoother and easier to find patterns of change later.

[0102] Step 4: Calculate the descent speed Find the slope (i.e., the first derivative) of the smoothed curve; that is, see how quickly the curve drops at each point. The more negative the slope, the more drastic the drop.

[0103] Step 5: Positioning the Stabilizing Wall On the entire slope curve, find the point where the coefficient of variation is most negative (i.e., the fastest decrease). The x-coordinate of this point is the location of the "stabilization wall." It represents the turning point where the square of the coefficient of variation decreases most rapidly as the mean increases.

[0104] Step 6: Assess the credibility (confidence interval) To determine the accuracy of this location, the original gene set is treated as a "bag," from which 500 random samples are repeatedly drawn (with replacement after each draw, simulating 500 experiments). Each time a sample is drawn, steps one through five are repeated, resulting in 500 stable wall locations. Finally, these 500 locations are arranged in ascending order, and the 2.5% and 97.5% positions are used as upper and lower bounds, forming a 95% confidence interval to represent the approximate range of fluctuation in the stable wall locations.

[0105] The working principle and technical effect of the above solution are as follows: This method plots all gene points in a bivariate distribution on a Cartesian coordinate system to form a bivariate scatter plot. After observing the pattern presented by the scatter plot, a two-segment continuous piecewise linear regression model is used to fit the data points, and the slope of the right segment (high expression region) is forced to be negative to reflect the biological law that the higher the expression level, the faster the noise decreases. The optimal parameters in the model are solved by nonlinear least squares method or global optimization algorithm, and the connection point of the two line segments is extracted, which is the location of the stability wall. To evaluate the statistical robustness of the stability wall location, a bootstrap resampling method is used: five hundred simulated datasets are generated by resampling with replacement from the original gene set. The above fitting process is repeated for each simulated dataset to obtain five hundred stability wall location estimates. The 0.025th quantile and 0.975th quantile of these estimates are calculated to obtain the 95% confidence interval of the stability wall location. At the same time, the absolute value of the right slope is calculated as the kurtosis, and the stability wall location, confidence interval, and kurtosis are combined to form complete stability wall information.

[0106] This method addresses the fundamental technical problem of the inability to objectively and automatically determine the boundary between the core and peripheral regions in existing technologies. Compared to existing methods that rely on subjectively set expression thresholds (e.g., greater than one per million transcripts) or physically meaningless quantiles (e.g., the top 10%), this method automatically identifies the inherent phase transition inflection points in the data through piecewise linear regression, ensuring that the boundary is determined by the data itself rather than by human intervention, significantly improving the objectivity and repeatability of the boundary division. By obtaining confidence intervals for the location of the stability wall through a bootstrap resampling method, it provides a statistical robustness assessment for the phase transition boundary for the first time, solving the problem of single-sample fitting results being affected by sampling fluctuations and effectively reducing the risk of misjudgment due to sample randomness. The introduction of kurtosis provides a new physical indicator for quantifying the urgency or criticality of system order, making it possible to quantitatively compare the speed of order establishment under different biological conditions.

[0107] In one embodiment of the present invention, obtaining 500 stable wall positions by performing 500 resampling operations with replacement on the gene set and repeating the above fitting process each time includes: The original gene set is obtained by treating all genes in the original gene set as a whole; By performing 500 independent resampling operations with replacement on the original gene set, and randomly selecting the same number of genes as the original gene set in each resampling, a simulated gene set of 500 is obtained. By repeatedly performing the complete fitting process for the stable wall localization on each simulated gene set, the estimated stable wall position value corresponding to each simulated gene set is obtained; By sequentially recording and aggregating the estimated stable wall locations corresponding to all 500 simulated gene sets, the values ​​of the 500 stable wall locations are obtained.

[0108] Specifically, by repeatedly performing the complete fitting process for stable wall localization on each simulated gene set, the estimated stable wall location for each simulated gene set is obtained, including: The current set of genes to be fitted is obtained by using one of the five hundred simulated gene sets as the current input dataset; A scatter plot of the current simulated dataset is obtained by constructing a bivariate scatter plot of the log mean and the square of the log coefficient of variation for all genes in the current simulated gene set. By fitting the scatter plot of the current simulated dataset with two consecutive piecewise linear regression models and constraining the slope of the right segment to a negative value, a piecewise linear regression model for the current simulated dataset is obtained. The optimal parameter combination for the current simulation dataset is obtained by solving the parameters in the piecewise linear regression model of the current simulation dataset using nonlinear least squares method or global optimization algorithm. By extracting the connection point of two line segments from the optimal parameter combination, the estimated location of the stable wall in the current simulation dataset is obtained. By repeating all the above operations sequentially for each of the five hundred simulated gene sets, the estimated location of the stable wall corresponding to each simulated gene set is obtained.

[0109] The working principle and technical effect of the above technical solution are as follows: This method treats all genes in the original gene set as a whole to form an original gene set. Five hundred independent resampling operations with replacement are performed on this original gene set. Each resampling randomly selects the same number of genes as the original gene set (allowing the same gene to be selected multiple times in a single sampling), generating five hundred simulated gene sets. For each simulated gene set, it is used as the current input dataset. A bivariate scatter plot of the logarithmic mean and logarithmic coefficient of variation squared is constructed for all genes in this set. Then, a two-segment continuous piecewise linear regression model is used for fitting, with the slope of the right segment constrained to be negative. The optimal parameter combination is solved using nonlinear least squares or a global optimization algorithm, and the connection point of the two line segments is extracted as the estimated value of the stable wall position for this simulated gene set. The above complete operation is repeated sequentially for each of the five hundred simulated gene sets, and all estimated values ​​are collected and recorded to obtain the values ​​of the five hundred stable wall positions.

[0110] This method addresses the problem in existing technologies where the lack of robustness assessment of phase transition boundaries prevents the determination of the consistency of stable wall locations across different samples or repeated experiments. By employing resampling with replacement, the process of multiple samplings from the original population is simulated, allowing the construction of a sampling distribution of stable wall locations. The choice of 500 resampling iterations strikes a balance between computational efficiency and estimation accuracy, ensuring the reliability of statistical estimation while avoiding excessive computational costs. Compared to a single-step fitting that only provides a single-point estimate, this method generates multiple simulated datasets and repeatedly fits them to obtain an empirical distribution of stable wall locations. This resampling-based robustness assessment method significantly improves the reliability and repeatability of stable wall location estimation, resulting in stronger consistency and credibility across different laboratories and batches of data.

[0111] In one embodiment of the present invention, S3 includes: The predicted logarithmic coefficient of variation squared is obtained by using the fitted piecewise linear model and the log mean of each gene. The residual for each gene is obtained by subtracting the predicted logarithmic coefficient of variation from the squared actual logarithmic coefficient of variation. The candidate gene set is obtained by comparing the log mean of each gene with the location of the stability wall and selecting genes whose log mean is strictly higher than the stability wall as candidates; The core region genes are obtained by judging the positive or negative value of the residual of each gene in the candidate gene set and retaining the genes with negative residuals. Genes in the outer region are obtained by classifying genes whose logarithmic mean is not higher than that of the stable wall and genes whose logarithmic mean is higher than that of the stable wall but whose residuals are positive into the outer region. The zero distribution of the residuals is obtained by repeatedly randomly shuffling the actual squared values ​​of the coefficient of variation and refitting the model each time. Core region genes are obtained by comparing the true residual of each gene with the 5th percentile of the zero distribution of residuals and determining that genes with true residuals less than the corresponding quantile are significant. By integrating the core region genes, an order core gene set is obtained.

[0112] The working principle and technical effect of the above solution are as follows: This method calculates the predicted logarithmic coefficient of variation (COP) squared value for each gene at a given expression level using a fitted piecewise linear model and the log-mean of each gene. The predicted value is subtracted from the actual COP squared value to obtain the residual for each gene. The sign of the residual reflects the relative level of the actual noise level of the gene compared to the model's expected level. The COP of each gene is compared with the location of the stability wall, and genes with a COP strictly higher than the stability wall are selected as candidate genes. The residuals of each gene in the candidate gene set are then evaluated for positive or negative values. Genes with negative residuals (i.e., actual noise lower than model prediction) are retained as core region genes. Genes with a COP not higher than the stability wall, and genes with a COP higher than the stability wall but positive residuals, are classified into the outer region. To further improve the reliability of core region genes, a permutation test was employed: the actual squared coefficients of variation were randomly shuffled multiple times, the model was refitted each time, and the residuals were calculated. A null distribution of the residuals was constructed, and the true residual of each gene was compared with the 5th percentile of this null distribution. Only genes with true residuals less than this quantile were retained as statistically significant core region genes. Finally, all core region genes that met the criteria were integrated to form an ordered core gene set.

[0113] This method addresses the technical problem of existing technologies failing to objectively distinguish between ordered core genes and disordered peripheral genes, and also overcomes the deficiency that existing technologies often identify correlational markers rather than causal targets. By simultaneously satisfying two conditions—expression levels above the stability wall and actual noise below model predictions—this method ensures that core region genes possess both sufficiently high expression intensity and sufficiently low expression fluctuations, truly representing the core molecular modules that maintain the basic order of cells. Compared to existing methods that only sort based on expression levels or only based on enrichment scores, this method's dual-condition screening strategy significantly improves the biological relevance and functional consistency of core region genes. The introduction of statistical significance judgment through permutation tests effectively eliminates false-positive genes that accidentally meet the screening conditions due to random fluctuations, significantly improving the reliability and reproducibility of the core region gene set. The explicit definition of the peripheral region provides a clear reference system for which genes enter the core region from the peripheral region during subsequent research to establish order.

[0114] In one embodiment of the present invention, the step of obtaining a candidate gene set by comparing the log-mean of each gene with the location of the stable wall and selecting genes whose log-mean is strictly higher than the stable wall as candidates includes: The log-mean value of each gene is obtained by using the fitted piecewise linear model and the log-mean of each gene. By comparing the log-mean of each gene with the location of the established stability wall, the relationship between the log-mean of each gene and the location of the stability wall can be obtained. An intermediate gene set is obtained by screening all genes for genes whose log mean is strictly greater than the location of the stable wall. By labeling and organizing each gene in the intermediate gene set, a candidate gene set is obtained.

[0115] Specifically, by screening all genes for those whose log-mean is strictly greater than the location of the stable wall, an intermediate gene set is obtained, including: The x-axis value of each gene is obtained by taking the logarithmic mean of all the genes obtained. The critical expression level values ​​for comparison are obtained by using the location of the established stability wall as a screening threshold. By comparing the log-mean value of each gene with the location of the stable wall, it is determined whether the log-mean value of each gene is strictly greater than the location of the stable wall, and the comparison result (greater than or not greater than) of each gene is obtained. By examining the comparison results of each gene in the entire gene pool one by one, all genes that meet the condition that the log mean is strictly greater than the position of the stable wall are screened out, thus obtaining a gene population that meets the condition of high expression level. By aggregating and organizing this gene population, an intermediate gene set is obtained, which represents all genes whose expression levels have crossed the phase transition boundary, serving as a candidate pool for further screening of core region genes.

[0116] The working principle and technical effects of the above solution are as follows: This method obtains the logarithmic mean value of each gene using a pre-fitted piecewise linear model and the logarithmic mean of each gene. The logarithmic mean of each gene is compared with the located stable wall position to determine whether the logarithmic mean of each gene is strictly greater than the stable wall position, obtaining the comparison result for each gene (greater than or not greater than). The comparison result of each gene is checked one by one from all genes, and all genes that meet the condition that the logarithmic mean is strictly greater than the stable wall position are selected, obtaining a gene population that meets the high expression level condition. This gene population is then aggregated and organized to form an intermediate gene set. This intermediate gene set represents all genes whose expression levels have crossed the phase transition boundary, serving as a candidate pool for further screening of core region genes.

[0117] This method addresses the technical problem of automated screening of highly expressed genes due to the lack of objective thresholds in existing technologies. Compared to existing methods that rely on manually set expression thresholds (e.g., greater than one per million transcripts) or dataset quantiles (e.g., the top 10% of expression levels), this method uses a stable wall position determined by the inherent phase transition law of the data as the screening threshold, giving the screening criteria biological and physical significance rather than statistical convenience. The stable wall position is the critical point where a qualitative change occurs from the outer region to the core region in the expression-noise joint distribution. Using this as the screening threshold ensures that the selected candidate genes have indeed biologically crossed the critical expression level required for order establishment. By defining the screening results as an intermediate gene set rather than the final result, the position of this step in the entire process is clarified: candidate pool generation rather than final judgment. This leaves room for fine-tuning by combining residuals, forming a hierarchical screening system and effectively reducing the false positive rate of the final core region gene set.

[0118] In one embodiment of the present invention, S4 includes: By using the core gene set as a reference, the expression vector of the core region gene in each sample is extracted to obtain the core region gene expression vector. The order parameter, ranging from zero to one, is obtained by calculating the ratio of the number of genes in the core region to the total number of genes after filtering; the first principal component value is obtained by performing principal component analysis on the expression matrix of genes in the core region in different samples and extracting the first principal component. By multiplying the first principal component value by negative one when it is opposite to the expected biological direction, the first principal component value after direction correction is obtained as the order parameter. The average expression order parameter is obtained by calculating the arithmetic mean of the normalized gene expression levels in the core region. The dynamic curve is obtained by calculating the above-mentioned order parameters and plotting them as curves from a series of samples collected along the time axis or the processing gradient. By observing the intervals where the order parameters in the dynamic curve undergo drastic changes, the phase transition interval is obtained, and thus the absolute cellular order parameters are obtained.

[0119] The working principle and technical effects of the above technical solution are as follows: This method uses the core gene set as a reference to extract the expression vector of the core region genes in each sample to be tested. One of three methods is used to calculate the order parameter: the ratio method calculates the ratio of the number of core region genes to the total number of filtered genes, obtaining a value between zero and one; the principal component analysis method performs principal component analysis on the expression matrix of core region genes in different samples, extracting the first principal component; if the direction of the first principal component is opposite to the expected biological direction, it is multiplied by negative one for correction; the average expression method calculates the arithmetic mean of the standardized expression levels of core region genes. For a series of samples collected along the time axis or processing gradient, the order parameter of each sample is calculated and plotted as a dynamic curve. By observing the intervals in the curve where the order parameter undergoes drastic changes, the phase transition interval of the system order is identified, thereby obtaining the absolute cellular order parameter.

[0120] This method addresses the fundamental technical problem in existing technologies where order indicators (such as first principal component score and enrichment score) are only relative values ​​and cannot be directly compared between different experiments. The order parameter provided by the proportional method has a strictly defined value range between zero and one, possessing clear physical meaning and representing the proportion of genes at the core of order. This allows for direct numerical comparison of order levels across different samples, species, and experimental conditions, achieving a leap from relative ranking to absolute measurement. Principal component analysis and average expression methods offer multiple optional computational paths to adapt to different application scenarios and research needs, enhancing the method's flexibility. Through dynamic curve plotting and identification of abrupt change intervals, it solves the problem that existing technologies cannot automatically identify critical windows during the establishment or collapse of order, making early prediction and intervention possible during cell differentiation, reprogramming, or pathological deterioration. This order parameter is highly sensitive to changes in cell state, capable of capturing minute fluctuations in order that traditional methods cannot detect, significantly improving the temporal resolution and early warning capabilities of cell state monitoring.

[0121] In one embodiment of the present invention, obtaining the phase transition interval by observing the interval where the order parameter in the dynamic curve undergoes a sharp change includes: By calculating the order parameter from a series of samples collected along the time axis or the processing gradient, the order parameter values ​​at multiple discrete time points or under processing conditions are obtained. By plotting the order parameter values ​​sequentially in chronological or gradient order, a dynamic curve of the order parameter changing with time or processing conditions can be obtained. By collecting information on the numerical trend of order parameters in dynamic curves, we can obtain overall morphological information on the curve's rising, falling, or plateauing phases. By identifying the intervals in the dynamic curve where the order parameter values ​​change drastically, the location and range of the jump intervals are obtained; by defining the jump intervals as the critical windows for the phase transition of the system order, the phase transition intervals are obtained.

[0122] Specifically, by identifying the intervals in the dynamic curve where the order parameter values ​​change drastically, the location and range of these intervals are obtained, including: By plotting the dynamic curve of the order parameter changing with time or processing conditions, a series of continuous data points on the curve and their corresponding order parameter values ​​are obtained. By calculating the magnitude of change in the order parameter values ​​between two adjacent data points on the dynamic curve, a sequence of changes between adjacent points is obtained. By calculating the average and standard deviation of the changes between all adjacent points on the entire dynamic curve, the statistical characteristics of the global changes can be obtained. By comparing the change between each adjacent point with the average of the global change plus twice the standard deviation, it is determined whether the change is significantly greater than the normal fluctuation range, thus obtaining the result of whether each interval belongs to a sharp jump. By filtering out all adjacent intervals that are determined to be abrupt changes from the dynamic curve, one or more candidate change intervals are obtained. By recording the time range or processing gradient range corresponding to each candidate transition interval as the specific location of the interval, and recording the distance from the start point to the end point of the interval as the range of the corresponding interval, the location and range of the transition interval are obtained. The transition interval marks the critical window when the system order undergoes a nonlinear phase transition.

[0123] The working principle and technical effect of the above technical solution are as follows: This method calculates the order parameter for a series of samples collected along the time axis or processing gradient, obtaining order parameter values ​​at multiple discrete time points or under different processing conditions. These parameter values ​​are then connected sequentially according to time or processing gradient to form a curve, obtaining a dynamic curve showing the change of the order parameter with time and processing conditions. The numerical trend information of the order parameter in the dynamic curve is collected to obtain the overall shape of the curve's rising, falling, or plateauing periods. When identifying abrupt change intervals, the change amplitude of the order parameter between two adjacent data points on the dynamic curve is first calculated to obtain a sequence of changes between adjacent points. Then, the average and standard deviation of the changes between all adjacent points on the entire dynamic curve are calculated to obtain the statistical characteristics of the global changes. The change between each adjacent point is compared with the average of the global changes plus twice the standard deviation to determine whether the change is significantly greater than the normal fluctuation range. All adjacent point intervals that are judged to be abruptly changed are filtered out to obtain candidate change intervals. The time range or processing gradient range corresponding to each candidate change interval is recorded as the specific location of the interval, and the distance from the start point to the end point of the interval is recorded as the range of the corresponding interval. Finally, the location and range of the change interval are obtained. This interval marks the critical window when the system order undergoes a nonlinear phase transition.

[0124] This method addresses the technical problem of existing technologies failing to automatically identify critical phase transition points during the establishment or collapse of order, and overcomes the limitation of existing continuous indicators (such as the first principal component score) which only provide monotonic trends and cannot identify nonlinear jumps. By plotting order parameter values ​​as dynamic curves in chronological order, the evolution of order is visualized, providing an intuitive foundation for subsequent quantitative analysis. By calculating the changes between adjacent points and comparing them with the statistical characteristics of global changes (mean plus two standard deviations), an objective mathematical definition of dramatic transitions is achieved, rather than relying on human observation or subjective judgment. This statistical threshold-based identification method can automatically distinguish between normal random fluctuations and true system phase transitions, effectively reducing the false alarm rate. The two standard deviation threshold corresponds to a statistically significant confidence level of approximately 95%, balancing sensitivity and specificity. The resulting phase transition intervals provide precise time windows for locating key turning points in cell differentiation, reprogramming, aging, or pathological deterioration.

[0125] According to one embodiment of the present invention, the system includes: The variable distribution module is used to calculate the mean and coefficient of variation square of the expression level of each gene in multiple samples or replicates by performing high-throughput transcriptome sequencing or gene chip detection on the samples, and obtain the bivariate distribution of the log mean and log CV² of each gene. The phase transition boundary module is used to obtain the phase transition boundary, i.e., the stability wall information and its confidence interval, by performing piecewise linear regression or sliding window fitting on genes through bivariate distribution. The automatic partitioning module is used to automatically partition genes into core and peripheral regions using stable wall information and fitting residuals to obtain an ordered core gene set. The parameter acquisition module is used to calculate the proportion of genes in the core region or the first principal component of a sample or time point using the core gene set of the order, thereby obtaining absolute cellular order parameters.

[0126] The working principle and technical effects of the above-mentioned technical solution are as follows: This method obtains raw gene expression data by collecting cell or tissue samples, extracting total ribonucleic acid, and performing high-throughput sequencing or gene chip detection. The arithmetic mean and standard deviation of the expression level of each gene in multiple samples or replicates are calculated, and then the coefficient of variation squared is calculated to quantify the degree of expression fluctuation (i.e., transcriptional noise) of each gene. After performing a base-10 logarithmic transformation on the mean expression level and the coefficient of variation squared, and adding a minimal constant, the logarithmic mean and logarithmic coefficient of variation squared for each gene are obtained, thus constructing a bivariate distribution for all genes. In this distribution, as the logarithmic mean increases, the logarithmic coefficient of variation squared shows a trend of first leveling off and then decreasing sharply. This invention automatically locates this inflection point of decrease using a piecewise linear regression model and defines it as the stability wall. Based on the location of the stability wall and the fitting residuals, genes with expression levels higher than the stability wall and actual noise lower than the model prediction are divided into the core region, and the remaining genes are divided into the peripheral region. By calculating the proportion of genes in the core region or performing principal component analysis on the gene expression matrix in the core region, absolute cellular order parameters that can be directly compared across samples and species can be obtained.

[0127] This method addresses the distortion in cell state characterization caused by existing techniques that rely solely on mean expression levels while completely ignoring the noise dimension. It also overcomes the technical deficiency of lacking objective, quantifiable phase transition boundaries to distinguish the core and periphery regions. Compared to existing techniques, this method significantly improves the accuracy and completeness of cell order assessment, transforming previously ignored noise dimensions into quantifiable order indicators. By introducing mathematically definable stability walls, it achieves a leap from subjective threshold-based classification to objective, data-driven classification, greatly reducing the bias and uncertainty introduced by manually set thresholds. The resulting absolute order parameters enable comparisons of cell states across different experiments, species, and sequencing platforms, effectively enhancing the method's universality and comparability.

[0128] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method of cell order parameter determination of a transcriptomic phase transition boundary, characterized in that, The method includes: S1. By performing high-throughput transcriptome sequencing or gene chip detection on the samples, the expression level of each gene in multiple samples or replicates is calculated using the mean and coefficient of variation squared to obtain the bivariate distribution of the log mean and log CV² of each gene. S2. By performing piecewise linear regression or sliding window fitting on the genes through bivariate distribution, the phase transition boundary, i.e., the stability wall information and its confidence interval, is obtained. S3. Automatically divide the genes into core and peripheral regions using stable wall information and fitting residuals to obtain an ordered core gene set; S4. Calculate the proportion of genes in the core region or the first principal component of the sample or time point using the core gene set of the order to obtain cellular order parameters.

2. The method for determining cellular order parameters at transcriptome phase transition boundaries according to claim 1, characterized in that, S1 includes: By collecting cell or tissue samples and extracting total ribonucleic acid, the samples are subjected to high-throughput sequencing or gene chip detection to obtain raw expression data; By comparison and quantification, the original expression level of each gene in multiple samples or replicates is quantified and standardized using the number of transcripts per million, the number of mapped readings per kilobase per million, or the original counting units to obtain the observed expression level. The expression mean and standard deviation are obtained by calculating the arithmetic mean and standard deviation of the expression levels of each gene in all samples or replicates. The coefficient of variation squared for each gene is obtained by dividing the standard deviation of each gene by the mean expression level and then squaring the result. By performing a base-10 logarithmic transformation on the mean expression level and the squared coefficient of variation of each gene, and adding a very small constant, the logarithmic mean and the logarithmic coefficient of variation of each gene are obtained. By combining the logarithmic mean and the squared logarithmic coefficient of variation of all genes, a bivariate distribution of the logarithmic mean and the squared logarithmic coefficient of variation is obtained.

3. The method for determining cellular order parameters at transcriptome phase transition boundaries according to claim 2, characterized in that, The method of obtaining a bivariate distribution of the logarithmic mean and the logarithmic coefficient of variation by combining the logarithmic mean and the logarithmic coefficient of variation of all genes includes: By combining the logarithmic mean and the squared logarithmic coefficient of variation of all genes in a one-to-one correspondence according to the gene index, a two-dimensional coordinate point corresponding to each gene is obtained. By pooling the two-dimensional coordinates of all genes into a single dataset, a bivariate distribution of the log mean and the log-squared coefficient of variation is obtained.

4. The method for determining cellular order parameters at transcriptome phase transition boundaries according to claim 1, characterized in that, S2 includes: A bivariate scatter plot is obtained by plotting the logarithmic mean and the squared logarithmic coefficient of variation of all genes in a Cartesian coordinate system. The bivariate distribution is fitted using a two-segment continuous piecewise linear regression model through a bivariate scatter plot, with the slope of the right segment constrained to be negative, thus obtaining the piecewise linear regression model. The location of the stable wall is obtained by solving the parameters in the piecewise linear regression model using nonlinear least squares method or global optimization algorithm; The location of the stable wall was obtained by smoothing and calculating the derivative of the bivariate distribution using a sliding window; the width of the sliding window was 0.3 log units (log 10 ), retrieve logs within each window 10 The local median of (CV²) is obtained by smoothing the median curve using a Savitzky-Golay filter (third-order polynomial, window length 51 points). The first derivative of the smoothed curve is calculated, and the minimum value of the first derivative (i.e. the point where CV² decreases the fastest as Mean increases) is taken as the location of the stabilizing wall x0. The 95% confidence interval is obtained by resampling (500 times) using Bootstrap. By resampling the gene set with replacement 500 times and repeating the above fitting process each time, the values ​​of 500 stable wall positions are obtained. By calculating the 0.025th and 0.975th quantiles of 500 stable wall locations, a 95% confidence interval for the stable wall locations is obtained. Steepness is obtained by calculating the absolute value of the slope on the right side; Complete information about the stable wall can be obtained by combining the location, confidence interval, and steepness of the stable wall.

5. The method for determining cellular order parameters at transcriptome phase transition boundaries according to claim 4, characterized in that, The process involves resampling the gene set with replacement 500 times, repeating the above fitting process each time, to obtain 500 values ​​for the stable wall positions, including: The original gene set is obtained by treating all genes in the original gene set as a whole; By performing 500 independent resampling operations with replacement on the original gene set, and randomly selecting the same number of genes as the original gene set in each resampling, a simulated gene set of 500 is obtained. By repeatedly performing the complete fitting process for the stable wall localization on each simulated gene set, the estimated stable wall position value corresponding to each simulated gene set is obtained; By sequentially recording and aggregating the estimated stable wall locations corresponding to all 500 simulated gene sets, the values ​​of the 500 stable wall locations are obtained.

6. The method for determining cellular order parameters at transcriptome phase transition boundaries according to claim 1, characterized in that, S3 includes: The predicted logarithmic coefficient of variation squared is obtained by using the fitted piecewise linear model and the log mean of each gene. The residual for each gene is obtained by subtracting the predicted logarithmic coefficient of variation from the squared actual logarithmic coefficient of variation. A candidate gene set is obtained by comparing the log mean of each gene with the location of the stability wall and selecting genes whose log mean is strictly higher than the stability wall as candidates; The core region genes are obtained by judging the positive or negative value of the residual of each gene in the candidate gene set and retaining the genes with negative residuals. Genes in the outer region are obtained by classifying genes whose logarithmic mean is not higher than that of the stable wall and genes whose logarithmic mean is higher than that of the stable wall but whose residuals are positive into the outer region. The zero distribution of the residuals is obtained by repeatedly randomly shuffling the actual squared values ​​of the coefficient of variation and refitting the model each time. Core region genes are obtained by comparing the true residual of each gene with the 5th percentile of the zero distribution of residuals and determining that genes with true residuals less than the corresponding quantile are significant. By integrating the core region genes, an order core gene set is obtained.

7. The method for determining cellular order parameters at transcriptome phase transition boundaries according to claim 6, characterized in that, The process involves comparing the log-mean of each gene with the location of the stable wall, and selecting genes whose log-mean is strictly higher than the stable wall as candidates to obtain a candidate gene set, including: The log-mean value of each gene is obtained by using the fitted piecewise linear model and the log-mean of each gene. By comparing the log-mean of each gene with the location of the established stability wall, the relationship between the log-mean of each gene and the location of the stability wall can be obtained. An intermediate gene set is obtained by screening all genes for genes whose log mean is strictly greater than the location of the stable wall. By labeling and organizing each gene in the intermediate gene set, a candidate gene set is obtained.

8. The method of claim 1, wherein, S4 includes: By using the core gene set as a reference, the expression vector of the core region gene in each sample is extracted to obtain the core region gene expression vector. The order parameter, ranging from zero to one, is obtained by calculating the ratio of the number of genes in the core region to the total number of genes after filtering; the first principal component value is obtained by performing principal component analysis on the expression matrix of genes in the core region in different samples and extracting the first principal component. By multiplying the first principal component value by negative one when it is opposite to the expected biological direction, the first principal component value after direction correction is obtained as the order parameter. The average expression order parameter is obtained by calculating the arithmetic mean of the normalized gene expression levels in the core region. The dynamic curve is obtained by calculating the above-mentioned order parameters and plotting them as curves from a series of samples collected along the time axis or the processing gradient. By observing the intervals where the order parameters in the dynamic curve undergo drastic changes, the phase transition interval is obtained, and thus the absolute cellular order parameters are obtained.

9. The method of claim 8, wherein the cell order parameter is determined by: ###0001### where N is the number of cells in the sample, and A is the area of the sample. The method of obtaining the phase transition interval by observing the interval where the order parameter in the dynamic curve undergoes a sharp change includes: ​ By calculating the order parameter from a series of samples collected along the time axis or processing gradient, the order parameter values ​​at multiple discrete time points or under processing conditions are obtained. By plotting the order parameter values ​​sequentially in chronological or gradient order, a dynamic curve of the order parameter changing with time or processing conditions can be obtained. By collecting information on the numerical trend of order parameters in dynamic curves, we can obtain overall morphological information on the curve's rising, falling, or plateau phases. By identifying the intervals in the dynamic curve where the order parameter values ​​change drastically, the location and range of the jump intervals are obtained; by defining the jump intervals as the critical windows for the phase transition of the system order, the phase transition intervals are obtained.

10. A system for implementing a method of cell order parameter determination of a transcriptomic phase transition boundary as claimed in claim 1, characterized in that, The system includes: The variable distribution module is used to calculate the mean and coefficient of variation square of the expression level of each gene in multiple samples or replicates by performing high-throughput transcriptome sequencing or gene chip detection on the samples, and obtain the bivariate distribution of the log mean and log CV² of each gene. The phase transition boundary module is used to obtain the phase transition boundary, i.e., the stability wall information and its confidence interval, by performing piecewise linear regression or sliding window fitting on genes through bivariate distribution. The automatic partitioning module is used to automatically partition genes into core and peripheral regions using stable wall information and fitting residuals to obtain an ordered core gene set. The parameter acquisition module is used to calculate the proportion of genes in the core region or the first principal component of a sample or time point using the core gene set of the order, thereby obtaining absolute cellular order parameters.