A response quantification method, device, medium and product of single-cell perturbation data

Abstract

The application discloses a single-cell perturbation data response quantification method, device, medium and product, relates to the field of single-cell sequencing data analysis, and comprises the following steps: obtaining a gene expression matrix and an initial perturbation label of a candidate perturbation cell; based on the gene expression matrix and the initial perturbation label, a cell similarity graph is constructed, a perturbation tendency score of each candidate perturbation cell is calculated according to the cell similarity graph by using a bivariate Moran's I correlation method, and a cell perturbation label is iteratively updated to obtain an updated perturbation label; the perturbation cells are weightedly sampled according to the updated perturbation label, the control cells are randomly sampled, and a plurality of sub-samples are generated; each sub-sample comprises a plurality of perturbation cells and a plurality of control cells; and closed-loop iterative evaluation is performed on the plurality of sub-samples until the gene response identity converges. The application can automatically correct inaccurate labels and technical noise, and can still stably and accurately quantify the perturbation response when the sample is unbalanced.

Description

Technical Field

[0001] This application relates to the field of single-cell sequencing data analysis, and in particular to a method, apparatus, medium, and product for quantifying the response of single-cell perturbation data. Background Technology

[0002] Cellular function can be systematically explored through perturbation experiments, including genetic perturbations (such as knockout, interference, or activation using CRISPR) and chemical perturbations (such as drugs or small molecules). In recent years, the combination of perturbation experiments with single-cell sequencing has given rise to high-resolution perturbation-measurement techniques (such as Perturb-seq and CROP-seq). The datasets generated by these techniques have grown exponentially, supporting a wide range of applications, such as elucidating gene regulation mechanisms, characterizing drug action mechanisms, and constructing virtual experimental surrogates. Therefore, accurately interpreting perturbation-driven transcriptional responses has become a crucial step in transforming these datasets into meaningful biological insights.

[0003] Data complexity continues to limit the interpretation of responses. At the experimental level, while perturbation tags (such as sgRNA and specific barcodes) can be obtained, the number of cells may be limited, and some cells may not exhibit a true response (perturbation escaping). Furthermore, technical noise generated during sample collection, culture, and sequencing exacerbates cell-to-cell non-independence. For example, batch effects, cell cycle, and library size can mask quantitative assessments of the response. At the analytical level, standard workflows for general single-cell data are not optimized for the uniqueness of perturbation data. Past practices often required combining specific algorithms at different stages for subsequent response quantification. In response quantification, linear regression strategies (such as MIMOSCA, scMAGeCk-LR, and Normalisr) are often only effective in certain steps, and workflows composed of multiple algorithms may struggle to adequately account for cell-to-cell non-independence in overall response quantification. This means that, when dealing with different datasets or different targets, combined workflows may still require additional specific designs to handle shared non-response factors between cells. Therefore, there is an urgent need for a method that can directly process noisy data, automatically correct inaccurate labels, and still stably and accurately quantify perturbation responses when samples are imbalanced. Summary of the Invention

[0004] The purpose of this application is to provide a method, device, medium, and product for quantifying the response of single-cell perturbation data, which can automatically correct inaccurate labels and still stably and accurately quantify the perturbation response when the samples are imbalanced.

[0005] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for response quantification of single-cell perturbation data, including: Obtain the gene expression matrix and initial perturbation tag for each candidate perturbation cell in the candidate perturbation cell group; the candidate perturbation cell group includes a few perturbation cells and several control cells; Based on the gene expression matrix and initial perturbation labels, a perturbation-prone cell similarity map is constructed. Using the bivariate Moran's I correlation method, the perturbation tendency score of each candidate perturbation cell is calculated according to the perturbation tendency cell similarity map. Based on the perturbation tendency score, the updated perturbation label is determined. Based on the updated perturbation label, perturbed cells are weighted and sampled, while control cells are randomly sampled to generate multiple subsamples; each subsample includes a number of perturbed cells and several control cells. A closed-loop iterative evaluation is performed on the multiple subsamples until the gene response identity converges, and the final gene response score and gene response identity are output. The closed-loop iterative evaluation includes: constructing a gene distribution cell similarity map for each subsample based on the current gene response identity, calculating the gene response score on the gene distribution cell similarity map; aggregating the gene response scores of all subsamples, updating the gene response identity of the subsamples, and reconstructing the gene distribution cell similarity map of the subsamples based on the updated gene response identity.

[0006] Secondly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the steps of the above-described response quantization method for single-cell perturbation data.

[0007] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described response quantization method for single-cell perturbation data.

[0008] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described response quantization method for single-cell perturbation data.

[0009] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, device, medium, and product for response quantification of single-cell perturbation data. It constructs a perturbation-prone cell similarity map based on a gene expression matrix and initial perturbation labels. Using the bivariate Moran's I correlation method, it calculates the perturbation tendency score for each candidate perturbation cell based on the similarity map. This score can be used to distinguish more likely real perturbation cells from spurious / mislabeled cells. Based on the updated perturbation labels using the perturbation tendency scores, perturbation cells are weighted and sampled to reduce the influence of mislabeled cells. This method can identify and correct labeling inaccuracies caused by spurious perturbation cells / perturbation escape / mislabeling, improving the reliability of the results. The method utilizes the bivariate Moran's I correlation method to construct the perturbation-prone cell similarity map based on the gene expression matrix and initial perturbation labels. The I-correlation method calculates the perturbation tendency score of each candidate perturbation cell based on the perturbation tendency cell similarity map, determines the updated perturbation label based on the perturbation tendency score, performs weighted sampling on perturbation cells based on the updated perturbation label, randomly samples control cells to generate multiple subsamples, performs closed-loop iterative evaluation on the multiple subsamples until the gene response identity converges, and can still stably output response results when the proportion of perturbation cells is low, the number of perturbation cells is small, or the control / perturbation sample is unbalanced. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is an application environment diagram of a response quantification method for single-cell perturbation data in one embodiment of this application.

[0012] Figure 2 This is a flowchart illustrating a method for response quantification of single-cell perturbation data provided in an embodiment of this application.

[0013] Figure 3 This is a detailed flowchart illustrating a method for quantifying the response of single-cell perturbation data, provided in an embodiment of this application.

[0014] Figure 4 This is a schematic diagram illustrating the principle of cell similarity map evaluation, provided in one embodiment of this application.

[0015] Figure 5 This is a schematic diagram illustrating the update of the perturbation trend score based on Moran's I correlation, provided as an embodiment of this application.

[0016] Figure 6 This is a schematic diagram of a disturbance-control comparison in an analysis unit provided in an embodiment of this application.

[0017] Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] The response quantification method for single-cell perturbation data provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server. Terminal 102 can send requests to be processed to server 104. Upon receiving the request, server 104 obtains the gene expression matrix and initial perturbation label of candidate perturbed cells. Based on the gene expression matrix and initial perturbation label, it constructs a cell similarity map, calculates the perturbation tendency score of each candidate perturbed cell using the bivariate Moran's I correlation method, iteratively updates the cell perturbation label, obtains the updated perturbation label, performs weighted sampling on candidate perturbed cells based on the updated perturbation label, randomly samples control cells to generate multiple subsamples, and performs closed-loop iterative evaluation on these subsamples until the gene response identity converges, obtaining the gene response evaluation result. Server 104 can feed back the obtained gene response evaluation result to terminal 102. In addition, in some embodiments, the response quantization method for single-cell perturbation data can also be implemented by the server 104 or the terminal 102 separately. For example, the terminal 102 can directly perform single-cell perturbation data response quantization for the request to be processed, or the server 104 can obtain the request to be processed from the data storage system and perform single-cell perturbation data response quantization for the request to be processed.

[0021] The terminal 102 can be, but is not limited to, various desktop computers and laptops. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers, or it can be a cloud server.

[0022] In one exemplary embodiment, such as Figure 2 As shown, a response quantization method for single-cell perturbation data is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 204.

[0023] Step 201: Obtain the gene expression matrix and initial perturbation tag for each candidate perturbation cell in the candidate perturbation cell group; the candidate perturbation cell group includes a few perturbation cells and several control cells.

[0024] Step 202: Based on the gene expression matrix and the initial perturbation label, construct a perturbation-prone cell similarity map, and use the bivariate Moran's I correlation method to calculate the perturbation tendency score of each candidate perturbation cell according to the perturbation-prone cell similarity map; based on the perturbation tendency score, determine the updated perturbation label.

[0025] Step 203: Weighted sampling of perturbation cells is performed according to the updated perturbation label, and random sampling of control cells is performed to generate multiple subsamples; each subsample includes a few perturbation cells and a number of control cells.

[0026] Step 204: Perform closed-loop iterative evaluation on the multiple subsamples until the gene response identity converges, and output the final gene response score and gene response identity. The closed-loop iterative evaluation includes: constructing a gene distribution cell similarity map for each subsample based on the current gene response identity, calculating the gene response score on the gene distribution cell similarity map; aggregating the gene response scores of all subsamples, updating the gene response identity of the subsamples, and reconstructing the gene distribution cell similarity map of the subsamples based on the updated gene response identity.

[0027] By implementing steps 201 to 204 above, a similarity map of perturbation-prone cells is constructed based on the gene expression matrix and initial perturbation labels. The perturbation tendency score of each candidate perturbation cell is calculated using the bivariate Moran's I correlation method based on the similarity map, and the cell perturbation labels are iteratively updated to obtain updated perturbation labels. Perturbation cells are then weighted and sampled according to the updated perturbation labels. This process can identify and correct labeling inaccuracies caused by "pseudo-perturbation cells / perturbation escape / mislabeling," improving the reliability of the results. The similarity map of perturbation-prone cells is constructed based on the gene expression matrix and initial perturbation labels, and the bivariate Moran's I correlation method is used to calculate the perturbation tendency score of each candidate perturbation cell. The I-correlation method calculates the perturbation tendency score of each candidate perturbation cell based on the perturbation tendency cell similarity map, and iteratively updates the cell perturbation label to obtain the updated perturbation label. The perturbation cells are weighted and sampled according to the updated perturbation label, and the control cells are randomly sampled to generate multiple subsamples. Closed-loop iterative evaluation is performed on the multiple subsamples until the gene response identity converges. Based on the above steps, this application can still stably output response results when the proportion of perturbation cells is low, the number of perturbation cells is small, the control / perturbation sample is unbalanced, or there is technical noise interference.

[0028] This application proposes CRANE (Closed-loop Response Analysis via Neighborhood-informed Evaluation): a framework for analyzing single-cell perturbation data based on Moran's I correlation, comprising steps 201-204 above. CRANE can handle various perturbation data scenarios, including data with imprecise labels (i.e., a high proportion of "pseudo-perturbation cells"). In CRANE, Moran's I correlation can be naturally viewed as a graph correction to Pearson correlation, thus helping to mitigate the impact of technical noise. Therefore, CRANE can directly use expression matrices contaminated with technical noise as input without additional external preprocessing. The core objective of CRANE is to quantify gene responses graphically. Response genes are expected to simultaneously satisfy two conditions: autocorrelation and bivariate correlation with perturbation identity; CRANE combines these two conditions using the L2 norm to calculate gene response scores. Furthermore, CRANE can be extended to functional gene set scoring, transforming noisy perturbation data into functional response knowledge that can be used for further interpretation.

[0029] The CRANE framework comprises three core steps: First, on a perturbation-prone cell similarity map, the perturbation tendency (CRANE-P score, i.e., perturbation tendency score) is quantified based on Moran's I correlation between each candidate perturbation cell and its perturbation identity. Second, multiple samples containing both perturbation and control cells are generated. For each sample, perturbation cells are sampled according to their perturbation tendency, while control cells are uniformly randomized. Third, a closed-loop iterative process is established. In each iteration, a gene distribution cell similarity map is first constructed within each sample, and Moran-based gene assessment is performed; then, response scores are integrated across samples, and gene response identities are updated, which in turn guide the map construction for the next iteration. Iteration stops when gene response identities stabilize, indicating that the alternating refinement between gene and cell assessments has reached equilibrium.

[0030] In this embodiment, each analysis unit corresponds to one perturbation-control comparison, such as... Figure 6 As shown. Typical inputs are a gene expression matrix and cell perturbation labels. Let... This represents the log-normalized gene expression matrix, where there are a total of One cell, measured One gene. Let The initial perturbation label of the cell, where =0 indicates a control group. =1 indicates a candidate perturbation, such as the detection of a specified sgRNA or drug treatment tag. Let Represents a cell similarity diagram, its elements w ij For cells With cells The similarity (connectivity) between them. Furthermore, let F = { f 1 ,f 2 , . . . ,f K} represents a set consisting of K functional genes, where each f k It is a subset of genes grouped according to their known biological functions. k =1, 2, 3, ... K .

[0031] The standard univariate form of Moran's I coefficient (autocorrelation), i.e., the autocorrelation Moran's I value, is defined as: (1); in, For variables The autocorrelation Moran's I value, For observation Target characteristics, For observation Target characteristics, and Variables and The mean, for and Spatial weights between observations, where N is the number of observations. Diagonal elements. Set it to 0.

[0032] The bivariate extended definition of Moran's I is as follows (2), that is, the formula for calculating the bivariate Moran's I value is: (2); in, For variables and The bivariate Moran's I value between cell characterization and the initial perturbation label is used in calculating the bivariate Moran's I value. For cell characterization, Initial perturbation label; For variables Observing candidate perturbed cells Target characteristics, For variables Observing candidate perturbed cells Target characteristics, For variables Observing candidate perturbed cells Target characteristics, For candidate perturbed cells With candidate perturbed cells Similarity between them The number of candidate perturbed cells; and Variables and The mean.

[0033] Moran's I applies not only to real-world physical coordinates but also to cell similarity maps derived from gene expression (such as the cell-cell connectivity matrix provided by the standard Scanpy workflow). The variables x and y in the above formulas can be cell-indexed vectors, including gene expression, perturbation labels, functional scores, or cell-level covariates.

[0034] CRANE uses Moran's I in a "graph structure-aware" manner to achieve stable and consistent gene assessment. CRANE considers a gene to be responsive when it exhibits consistency with both the cellular graph structure and the perturbation identity information. For each gene, CRANE combines its Moran's I autocorrelation (Equation (1)) and its bivariate Moran's I with the perturbation identity information (Equation (2)) by using their L2 norm to form a single gene response score (RS).

[0035] Furthermore, in CRANE, Moran's I can be interpreted as a Pearson correlation corrected by the cell similarity map W, thereby reducing technical noise while improving the ability to capture the response.

[0036] The classic Pearson correlation coefficient is defined as: (3); in, For variables and The Pearson correlation coefficient between them.

[0037] Formula (2) can be rewritten as an equivalent form that is more similar to formula (3): (4); Among them, molecules ( This can be viewed as using normalized weights. The weighted covariance. This mechanism can improve response capture capability while reducing the impact of technical noise in an inherently self-correcting manner, and can also be extended to evaluate other cell-indexed input data.

[0038] After obtaining the gene expression matrix and initial perturbation labels, such as Figure 3 As shown, the process includes three steps: first, in step 1, the perturbation tendency of each cell is updated to exclude perturbation escape cells; second, in step 2, weighted sampling is performed based on the perturbation tendency to generate multiple control-perturbation balanced samples; and third, in step 3, a closed-loop iteration of gene evaluation and cell similarity map construction is performed across sampling to finally obtain stable and consistent gene response evaluation results.

[0039] The gene response assessment results include: 1) the cell perturbation tendency score (CRANE-P score), which indicates whether each cell is perturbed or escapes. Values ​​range from -1 to 1, with a value greater than 0 indicating a perturbation tendency. 2) the gene response score (…). ), as a quantitative value for response considering cell non-independence, takes a value of 0-1, with a larger value indicating a higher degree of response. 3) Gene response identity ( The binary vector () is used to indicate whether a response has occurred. It serves both as the result of identifying the biologically significant set of response genes and as a means to build a cell similarity map to assist in the evaluation of Moran's I response of other cell index vectors.

[0040] Step 1 (corresponding to step 202): Update cell perturbation tendency: A dataset containing cell perturbation labels is obtained using cell sequencing technology. Candidate perturbation cells do not necessarily exhibit true perturbation responses, although their initial perturbation label z can still provide a rough indication. Guided by the initial perturbation label z, a cell similarity map is constructed, and the perturbation tendency of each candidate perturbation cell is quantified, called the perturbation tendency score (e.g., ...). Figure 5 (As shown).

[0041] First, a similarity map of perturbation-prone cells was constructed following these steps. To better capture the nearest neighbor relationships between perturbation cells and control cells, a Kolmogorov–Smirnov (KS) test was performed based on the initial perturbation label z, and the cells with the smallest p-value were selected. One gene (optional) =100). Based on these genes, the gene expression matrix E is truncated to obtain the truncated expression matrix. Then input the scanpy process. Run the sc.pp.neighbors function (with k=20 optional) using cosine distance to compute the truncated representation matrix. The connectivity matrix W′ is obtained and used as a cell similarity graph.

[0042] Secondly, the perturbation tendency of each candidate perturbed cell is quantified using bivariate Moran's I. The rationale is that cells similar to perturbed cells are more likely to also be in a perturbed state. Since Moran's I requires a vector by cell index as input, the cell representation of each cell c is represented by the cosine similarity vector Sc relative to all cells. The cosine similarity vector Sc is based on a truncated expression matrix. Calculated using sklearn. Using formula (2), in the connectivity matrix... Substitute Sc into the variable in formula (2) Substitute the disturbance label z into the variable in formula (2). Calculate the bivariate Moran's I value to obtain I. cz The obtained I czThe perturbation tendency score for cell c is used to quantify the consistency between the cell's similarity pattern and the perturbation label z. A perturbation tendency score close to +1 indicates a responsive cell, while a perturbation tendency score close to -1 indicates a non-responsive cell. z′ = max(0, I cz )·z c The formula updates the perturbation label, resulting in the updated perturbation label. Updated perturbation labels Substitute the above process into the recalculation of the perturbation tendency score I. cz And update the perturbation label itself; when the bivariate Moran's I value between the updated perturbation labels z′ of two adjacent iterations (substitute the updated perturbation labels z′ of two adjacent iterations into the variables in formula (2) respectively) and variables The iteration stops when the bivariate Moran's I value between the updated perturbation labels z′ of two consecutive iterations exceeds a preset threshold. The preset threshold can be 0.9.

[0043] In step 202, based on the gene expression matrix and the initial perturbation label, a perturbation-prone cell similarity map is constructed. Using the bivariate Moran's I correlation method, the perturbation tendency score of each candidate perturbation cell is calculated based on the perturbation-prone cell similarity map. Specifically, this includes: Based on the initial perturbation labels, the gene expression matrix was tested for differences, and a predetermined number of genes with the highest significant differences were selected to construct a truncated expression matrix. Based on the truncated expression matrix, the cosine distance between cells is calculated, a connectivity matrix is ​​constructed, and the connectivity matrix is ​​determined as a similarity map of cells prone to perturbation. For each candidate perturbation cell, the cosine similarity vector of the candidate perturbation cell relative to all candidate perturbation cells is calculated as the cell representation. The bivariate Moran's I value of the cell representation and the initial perturbation label is calculated on the connectivity matrix as the perturbation tendency score of the candidate perturbation cell.

[0044] In step 202, based on the perturbation tendency score, an updated perturbation label is determined, specifically including: updating the perturbation label using the perturbation tendency score; iteratively recalculating the bivariate Moran's I value with the updated perturbation label until the bivariate Moran's I value between the updated perturbation labels obtained after two adjacent iterations exceeds a preset threshold, at which point the iteration stops.

[0045] Step 2 (corresponding to step 203 above), multiple weighted sampling: To address the imbalance between the number of perturbed cells and control cells, multiple weighted sampling is used instead of rebinding the labels to hard 0 / 1. Control cells are obtained through random sampling, while candidate perturbed cells are selected based on the updated perturbed labels obtained in step 202. After performing softmax normalization, the updated perturbation label is then... Weights are assigned for weighted sampling. This process generates p samples ( Figure 3 In this model, subsample 1 corresponds to subsample 1, subsample 2 corresponds to subsample 2, and subsample p corresponds to subsample p). Each subsample contains q control cells and q perturbation cells, with p=5 and q=50. This strategy ensures that even with few perturbation cells, each subsample can still mix different control backgrounds with perturbation cells. Aggregating the results from multiple subsamples reduces bias from a single sample and improves the stability of response assessment.

[0046] Step 3 (corresponding to step 204 above), closed-loop optimization between gene response assessment and cell similarity map: CRANE runs a closed-loop iterative process in the multi-sampling settings generated in step 203 to achieve alternating optimization between gene response assessment and cell similarity map (W). The goal is to obtain stable and consistent gene evaluation, ultimately outputting gene response scores (RS, response indicators based on Moran's I) and gene response identities (RI, binary response indicators). The steps of the closed-loop iterative assessment are as follows: Step 3.1, Closed-loop design and optimization objectives: Let t denote the iterative index, then the process can be written as: Since the construction of the cell similarity map (W) depends on the selection of responding genes (represented by gene response identity RI), this closed loop can be further expanded as follows: These components form a closed loop, but direct joint optimization of W and RS is quite difficult. Therefore, we will start from... arrive The intermediate mapping is regarded as a function This simplifies the closed loop to: This rewrite transforms the original continuous problem into a discrete optimization task. It uses Gene Response Identity (RI) as the direct optimization objective, aiming to find a solution in the discrete state space that satisfies... The fixed point. It's important to note that CRANE does not allow each subsample to converge independently before aggregation. At iteration t, all samples share the same gene response identity. The mapping f is performed within each sample, and then the evaluations of each subsample are aggregated to obtain the result. When RI converges, the corresponding cell similarity map W and gene response score RS also tend to stabilize.

[0047] The CRANE iteration is formulated as a feature selection task in a discrete state space. To simplify optimization and avoid oscillations, a reverse elimination process similar to recursive feature elimination (RFE) is employed. This unidirectional design progressively removes non-responsive genes (setting RI to 0) and retains responsive genes (keeping RI at 1). Allowing initialization from a high-recall gene set RI0, the iterations refine the selection results progressively by removing false positives.

[0048] Each sample receives the genetic response identity from the previous iteration. Based on this, a cell similarity map is constructed to generate a Moran-based gene response assessment. First, each sample is identified based on its gene response identity from the previous iteration. The gene expression matrix is ​​subsetted, and a cell similarity graph is constructed. This cell similarity graph is then symmetricized and converted into a connectivity matrix using Scanpy. W t Secondly, for a given gene g, in the current cell similarity graph... W t Moran's I statistic is calculated. Specifically, its autocorrelation Moran's I value is calculated using formula (1). g And the gene expression matrix of the gene and the bivariate Moran's I value of the perturbation tag were calculated using formula (2). In formula (2) and y Corresponding to gene expression matrix E g With the perturbation label z. Generally, CRANE considers a gene as a responding gene when it simultaneously exhibits behavior consistent with both the graph structure and the perturbation label. Therefore, the two correlations are aggregated, namely, the autocorrelation Moran's I value and the bivariate Moran's I value are aggregated into and using non-responsive genes ( Centering the median calculated at (=0) yields the gene response score. The principle of cell similarity map assessment is as follows: Figure 4 As shown. That is, the gene response score of the gene. = .

[0049] In step 204, a gene distribution cell similarity map is constructed for each subsample based on the current gene response identity, and a gene response score is calculated on the gene distribution cell similarity map. This specifically includes the following steps: Based on the gene response identities obtained in the previous iteration, a subset of response genes is selected from the gene expression matrix to construct a gene distribution cell similarity map for the current iteration. Within each subsample, based on the current gene distribution cell similarity map, calculate the autocorrelation Moran's I value of each gene and the bivariate Moran's I value of the gene expression matrix and perturbation label. The autocorrelation Moran's I value and the bivariate Moran's I value are combined to form the original response score, which is then centered by the median of the gene response scores of non-responding genes to obtain the gene response scores within each subsample.

[0050] Step 3.4, Cross-sample comprehensive evaluation response score vector: gene response score of subsample Gene response score of each gene g The average is obtained by taking the average across samples. Genes marked as non-responsive in the previous iteration ( =0) as a background reference, response threshold Set it to the median of its scores plus twice the standard deviation. Then update the response identity vector as follows: ,in This indicates the indicator function. This update implements the reverse elimination process for the aforementioned RFE class.

[0051] Step 3.5, Convergence Design and Criteria: CRANE updates the response identity vector in a discrete and finite state space. This update is monotonic, thus guaranteeing convergence to a fixed point. At the fixed point, further iterations do not change the solution, thus avoiding overfitting in the mechanism. To detect convergence, the algorithm stops when the Jaccard distance between adjacent RI vectors remains 0 for four consecutive iterations. The maximum number of iterations is set to 50. CRANE further stabilizes the iterative process by monitoring the reverse elimination rate derived from the Jaccard distance.

[0052] In step 204, closed-loop iterative evaluation is performed on the multiple sub-samples until the gene response identity converges, and the final gene response score and gene response identity are output, specifically including: The aggregate response score is obtained by averaging the gene response scores of all subsamples. Using the statistical distribution of non-response gene scores as the response threshold, the aggregated response scores are converted into binary gene response identities. Reconstruct the gene distribution cell similarity map of the subsamples based on the updated gene response identity; The iteration is considered converged and terminated when the Jaccard distance between the gene response identities after a set number of consecutive iterations reaches zero. The set number of iterations can be 4.

[0053] In a specific example, the response quantification method for single-cell perturbation data further includes an extension to a functional gene set assessment process, which includes the following steps: After obtaining the converged gene response identities, an aggregated cell similarity map is constructed based on the converged gene response identities and all subsamples. For a functional gene set containing multiple genes, the gene expression matrix of the functional gene set is extracted and decomposed by principal component analysis (PCA) to obtain multiple principal component vectors indexed by cell. For each principal component vector, calculate the autocorrelation Moran's I value of the principal component vector and the bivariate Moran's I value of the principal component vector and the perturbation label, and calculate the gene response score of each principal component vector based on the autocorrelation Moran's I value and the bivariate Moran's I value of the principal component vector. The principal component vector with the highest gene response score is selected as the representative of the functional gene set's response, and the response quantification results at the functional gene set level are output.

[0054] After convergence, the identity of the gene response based on convergence. And the balanced cell population obtained in step 203 (the balanced cell population consists of all subsamples), construct an aggregated cell similarity map. This cell similarity graph serves as a stable graph structure basis, allowing Moran's I (Equations (1) and (2)) to be used to evaluate additional cell-indexed vectors. Here, we focus on its application in scoring functional gene sets.

[0055] CRANE employs a principal component analysis-based decomposition strategy to represent functional inputs. This approach performs well in benchmark evaluations on standard single-cell data. For data containing M... k Functional gene set of one gene Extract its gene expression matrix This is then projected onto the principal component vectors (PCs). These PCs are treated as cell-indexed vectors and may capture the response signal or technical noise. In a specific example, consider log3M. k Each principal component vector is evaluated using the same Moran's I-based RS index as individual genes. Unlike traditional methods, CRANE does not presuppose whether a principal component vector represents a response or noise; each principal component vector is then evaluated using the same RS index as individual genes. The principal component vectors with the highest scores are then used as the functional gene set. The representative outputs the quantitative results of the response at the functional gene set level.

[0056] This application proposes the CRANE framework. This method breaks through the traditional assumption of "independent cells," explicitly treating cells as non-independent individuals and combining cell-level relationships (similarity graphs) with gene-level response quantification, using Moran's I correlation index as its core implementation. Therefore, the CRANE method combines flexibility and robustness when dealing with the complexity of perturbation data. First, as a non-parametric method, CRANE does not require modeling of data distribution, thus it is applicable to various data types and perturbation conditions. Second, CRANE does not rely on additional perturbation cell screening or technical noise preprocessing steps; even with a small number of perturbation cells, it can adaptively reduce the impact of technical noise on response evaluation through a closed-loop iterative mechanism.

[0057] In this application, a systematic benchmark was constructed across diverse biological backgrounds (including clinical samples) and perturbation types, demonstrating CRANE's applicability to perturbation data. Furthermore, single-cell perturbation data (base editing) were generated for key genes in the MAPK pathway, validating CRANE's scalability. The results indicate that CRANE can also integrate external knowledge resources to assess the response of functional gene set scores. Overall, CRANE provides an adaptive response quantification strategy that simplifies complex perturbation analysis processes and improves the interpretability of results.

[0058] Compared to common single-cell perturbation response quantification methods in related technologies (such as those mainly relying on differential expression / LogFC and its variants, relying on pre-filtering of spurious perturbation cells, or strongly relying on additional denoising / regression preprocessing), the response quantification method proposed in this application, "based on the correlation between cell similarity maps and Moran's I (including bivariate Moran's I)," has the following technical effects and advantages: (1) It can identify and correct the labeling problems caused by “pseudo-perturbation cells / perturbation escape / mislabeling”, and improve the reliability of the results.

[0059] Source technical points: Part 2 "Update cell perturbation tendency (step 202, CRANE-P)" and "Tendency-based weighted sampling (step 203)".

[0060] Causes / Mechanisms: Existing methods typically assume that the given perturbation labels are completely correct. However, in real experiments, it is common for cells to be labeled as perturbations but fail to produce an effective biological response (perturbation escape), or for mislabeling to occur due to label noise. This method evaluates the consistency between "candidate perturbation cells and perturbation identities" on a cell similarity map to obtain a perturbation propensity score. This score can be used to distinguish more likely real perturbation cells from spurious perturbation / mislabeled cells, and to reduce the influence of mislabeled cells in subsequent sampling and scoring processes. If necessary, it can also be used to update / correct the original binary perturbation labels, thereby improving the reliability of downstream response quantification.

[0061] (2) It can still output stable response results when the proportion of perturbed cells is low, the number of perturbed cells is small, or the control / perturbed sample is unbalanced.

[0062] Source technical points: Part 2 "Update cell perturbation tendency (step 202)" + "Multiple weighted sampling / balanced sampling (step 203)" + "Closed-loop iterative optimization (step 204)".

[0063] Causes / Mechanisms: Existing processes often suffer from unstable estimations and highly sensitive results to sampling when there are few or a low proportion of perturbed cells. This application first calculates the perturbing tendency (reflecting the degree to which it is more likely to be a true perturbed cell) for each candidate perturbed cell, and then performs multiple rounds of weighted sampling based on this, so that each round of analysis is performed on a more balanced and information-rich cell set; at the same time, in the closed-loop iteration, the "current more credible cell set" is continuously used to optimize the graph structure and response gene set, thereby significantly improving the stability and usability in small sample / imbalanced scenarios.

[0064] (3) The response quantization accuracy is higher, and it can more reliably distinguish between the real disturbance response and the non-response / noise.

[0065] Source technical points: Part 2, “Calculate Moran’s I (including bivariate Moran’s I) correlation on the cell similarity map and define the gene response score RS / correlation index accordingly” and “Closed-loop iterative optimization between gene response assessment and cell similarity map (step 204)”.

[0066] Causes / Mechanisms: Existing methods often quantify responses directly at the cellular level using expression differences or simple statistics, which are easily confused by sequencing depth, batch effects, technical noise, or differences in cell state. This application places the calculation of "response correlation" on a cell similarity map and uses Moran's I, a map-aware correlation metric, making the scoring more focused on "consistent change patterns within similar cell neighborhoods," thereby reducing spurious correlations caused by random fluctuations or local noise and improving the accuracy and stability of true response identification.

[0067] (4) It has stronger resistance to technical noise and can complete response quantization without relying on additional complex preprocessing, reducing labor / process costs.

[0068] Source technical points: Part 2 "Correlation calculation of Moran's I (including bivariate Moran's I) based on cell similarity graphs" and "Alternating optimization of graph structure and response gene set in closed-loop iteration (step 204)".

[0069] Causes / Mechanisms: Many related technologies require additional denoising, regression correction, or strong-dependency pre-filtering strategies; otherwise, technical noise can easily be mistaken for biological responses. This application uses a correlation metric constrained by a graph structure and gradually weakens the dominant influence of noisy genes or unstable associations on the graph structure through closed-loop iteration. This allows the method to directly process noisy expression matrix inputs, reducing the dependence on additional preprocessing while still obtaining stable and interpretable response scores (gene response scores), thereby lowering the barrier to entry and the cost of manual parameter tuning.

[0070] This application also provides an application scenario in which the above-mentioned response quantification method for single-cell perturbation data is applied. Specifically, the response quantification method for single-cell perturbation data provided in this embodiment can be applied in a drug action mechanism characterization scenario. The drug action mechanism characterization scenario includes a content production stage, a single-cell perturbation data response quantification link, and a drug action mechanism characterization stage. Gene expression matrices and initial perturbation tags enter the single-cell perturbation data response quantification link from the content production stage, obtain corresponding gene response evaluation results through human-machine collaboration, and then enter the downstream drug action mechanism characterization stage. The response quantification method for single-cell perturbation data provided in this embodiment belongs to the single-cell perturbation data response quantification link. Specifically, in the single-cell perturbation data response quantification process targeting gene expression matrices and initial perturbation labels, the gene expression matrices and initial perturbation labels of candidate perturbation cells can be obtained. Based on the gene expression matrices and initial perturbation labels, a cell similarity map is constructed. The bivariate Moran's I correlation method is used to calculate the perturbation tendency score of each candidate perturbation cell according to the cell similarity map, and the cell perturbation labels are iteratively updated to obtain updated perturbation labels. Based on the updated perturbation labels, candidate perturbation cells are weighted and sampled, and control cells are randomly sampled to generate multiple subsamples. Closed-loop iterative evaluation is performed on the multiple subsamples until the gene response identity converges, and the gene response evaluation result is obtained.

[0071] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores cell perturbation response quantization data. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a single-cell perturbation data response quantization method.

[0072] Those skilled in the art will understand that Figure 7 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0073] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0074] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0075] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of the relevant data are carried out in compliance with the relevant data protection laws and policies of the country where the location is located, and with the authorization granted by the owner of the corresponding device.

[0076] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0077] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0078] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0079] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for response quantization of single-cell perturbation data, characterized in that, The response quantification method for the single-cell perturbation data includes: Obtain the gene expression matrix and initial perturbation tag for each candidate perturbation cell in the candidate perturbation cell group; the candidate perturbation cell group includes a few perturbation cells and several control cells; Based on the gene expression matrix and initial perturbation labels, a perturbation-prone cell similarity map is constructed. Using the bivariate Moran's I correlation method, the perturbation tendency score of each candidate perturbation cell is calculated according to the perturbation tendency cell similarity map. Based on the perturbation tendency score, the updated perturbation label is determined. Based on the updated perturbation label, perturbed cells are weighted and sampled, while control cells are randomly sampled to generate multiple subsamples; each subsample includes a number of perturbed cells and several control cells. A closed-loop iterative evaluation is performed on the multiple subsamples until the gene response identity converges, and the final gene response score and gene response identity are output. The closed-loop iterative evaluation includes: constructing a gene distribution cell similarity map for each subsample based on the current gene response identity, calculating the gene response score on the gene distribution cell similarity map; aggregating the gene response scores of all subsamples, updating the gene response identity of the subsamples, and reconstructing the gene distribution cell similarity map of the subsamples based on the updated gene response identity.

2. The response quantization method for single-cell perturbation data according to claim 1, characterized in that, Based on the gene expression matrix and initial perturbation labels, a perturbation-prone cell similarity map is constructed. Using the bivariate Moran's I correlation method, the perturbation tendency score of each candidate perturbation cell is calculated based on the perturbation-prone cell similarity map, specifically including: Based on the initial perturbation labels, the gene expression matrix was tested for differences, and a predetermined number of genes with the highest significant differences were selected to construct a truncated expression matrix. Based on the truncated expression matrix, the cosine distance between cells is calculated, a connectivity matrix is ​​constructed, and the connectivity matrix is ​​determined as a similarity map of cells prone to perturbation. For each candidate perturbation cell, the cosine similarity vector of the candidate perturbation cell relative to all candidate perturbation cells is calculated as the cell representation. The bivariate Moran's I value of the cell representation and the initial perturbation label is calculated on the connectivity matrix as the perturbation tendency score of the candidate perturbation cell.

3. The response quantification method for single-cell perturbation data according to claim 2, characterized in that, The formula for calculating the bivariate Moran's I value is: ； in, For variables and The bivariate Moran's I value between cell characterization and the initial perturbation label is used in calculating the bivariate Moran's I value. For cell characterization, Initial perturbation label; For variables Observing candidate perturbed cells Target characteristics, For variables Observing candidate perturbed cells Target characteristics, For variables Observing candidate perturbed cells Target characteristics, For candidate perturbed cells With candidate perturbed cells Similarity between them The number of candidate perturbed cells; and Variables and The mean.

4. The response quantization method for single-cell perturbation data according to claim 1, characterized in that, Based on the disturbance tendency score, the updated disturbance label is determined, specifically including: Update the perturbation label using the perturbation tendency score; The bivariate Moran's I value is recalculated iteratively with the updated perturbation label until the bivariate Moran's I value between the updated perturbation labels obtained after two adjacent iterations exceeds a preset threshold, at which point the iteration stops.

5. The response quantization method for single-cell perturbation data according to claim 1, characterized in that, Based on the current gene response identity, a gene distribution cell similarity map is constructed for each subsample. A gene response score is then calculated on this gene distribution cell similarity map, specifically including: Based on the gene response identities obtained in the previous iteration, a subset of response genes is selected from the gene expression matrix to construct a gene distribution cell similarity map for the current iteration. Within each subsample, based on the current gene distribution cell similarity map, calculate the autocorrelation Moran's I value of each gene and the bivariate Moran's I value of the gene expression matrix and perturbation label. The autocorrelation Moran's I value and the bivariate Moran's I value are combined to form the original response score, which is then centered by the median of the gene response scores of non-responding genes to obtain the gene response scores within each subsample.

6. The response quantization method for single-cell perturbation data according to claim 1, characterized in that, Perform closed-loop iterative evaluation on the multiple subsamples until the gene response identity converges, and output the final gene response score and gene response identity, specifically including: The aggregate response score is obtained by averaging the gene response scores of all subsamples. Using the statistical distribution of non-response gene scores as the response threshold, the aggregated response scores are converted into binary gene response identities. Reconstruct the gene distribution cell similarity map of the subsamples based on the updated gene response identity; When the Jaccard distance between the gene response identities of consecutive iterations for a set number of iterations is zero, convergence is determined and the iteration is terminated.

7. The response quantization method for single-cell perturbation data according to claim 1, characterized in that, The response quantification method for single-cell perturbation data also includes: After obtaining the converged gene response identities, an aggregated cell similarity map is constructed based on the converged gene response identities and all subsamples. For a functional gene set containing multiple genes, the gene expression matrix of the functional gene set is extracted and decomposed by principal component analysis to obtain multiple principal component vectors indexed by cell. For each principal component vector, calculate the autocorrelation Moran's I value of the principal component vector and the bivariate Moran's I value of the principal component vector and the perturbation label, and calculate the gene response score of each principal component vector based on the autocorrelation Moran's I value and the bivariate Moran's I value of the principal component vector. The principal component vector with the highest gene response score is selected as the representative of the functional gene set's response, and the response quantification results at the functional gene set level are output.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement the response quantization method for single-cell perturbation data according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the response quantization method for single-cell perturbation data as described in any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the response quantization method for single-cell perturbation data as described in any one of claims 1-7.

Images