Space-time immune score method and its application in efficacy prediction of anti-tumor immunotherapy

Abstract

The application provides a spatiotemporal immune scoring method and application thereof in antitumor immunotherapy efficacy prediction, and the scoring method comprises the following steps: constructing an original four-dimensional data matrix; constructing a time dynamic model for each individual number-immune cell type combination of the four-dimensional data, to obtain a time trend parameter; based on the four-dimensional tensor of the sampling time point at the beginning and end of treatment, using the time dynamic model to calculate the relative change rate of each immune cell type during the treatment, to obtain a cumulative change parameter; calculating the standardized effect size between the treatment sensitive group and the treatment resistant group at each sampling time point for each immune cell type, to obtain an effect size parameter; based on the time trend parameter, the cumulative change parameter and the effect size parameter, a standardized multidimensional weight coefficient is calculated, and then the time dynamic model is combined to calculate a spatiotemporal joint score. The application can solve the problems of poor early prediction effect and low precision when a single biomarker is used for immunotherapy efficacy prediction.

Description

Technical Field

[0001] This invention relates to the field of biomedical informatics combined with immunotherapy efficacy prediction technology, specifically to a spatiotemporal immune scoring method and its application in predicting the efficacy of antitumor immunotherapy. Background Technology

[0002] Immune checkpoint inhibitors (ICIs) have made groundbreaking progress in the treatment of various solid tumors. However, in tumors such as colorectal cancer, only a few subtypes, such as microsatellite highly unstable (MSI-H), can significantly benefit from them, while the vast majority of microsatellite stable (MSS) patients have low response rates. How to accurately predict patient response before or early in treatment, thereby guiding clinical decisions and avoiding ineffective treatment and wasted resources, is a key technical problem that urgently needs to be solved.

[0003] Existing technologies for predicting treatment efficacy, including tumor mutational burden (TMB), microsatellite instability, PD-L1 expression levels, and immune cell scoring systems such as Immunoscore, TIDE score, Immunoscore-IC, and ICP score, generally suffer from the following technical deficiencies: They employ static, single-time-point assessments, with most indicators only sampled and calculated at baseline or a fixed time point before treatment, failing to capture the dynamic evolution of immune status during treatment; they use a single spatial dimension for assessment, with most methods focusing only on one sampling site in the tumor microenvironment (TME) or peripheral blood, failing to integrate information from multiple sites and thus struggling to reflect the synergistic changes in systemic and local immune status; cell type coverage is limited: some scoring systems only involve T cells or a limited number of immune cell types, failing to cover cell populations such as myeloid suppressor cells and dendritic cells that play important roles in drug resistance mechanisms; and there is a lack of weight construction and standardization methods that can be validated across species and cohorts, resulting in insufficient predictive efficacy and stability of the models in independent validation sets or human samples.

[0004] Therefore, there is an urgent need for a mathematical modeling method that can integrate dynamic change information from multiple time points, multiple spatial sites (peripheral blood and tumor tissue), and multiple immune cell types, to construct a quantifiable, standardized immunotherapy efficacy prediction scoring system with cross-cohort / cross-species migration capabilities. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a spatiotemporal immune scoring method and its application in predicting the efficacy of antitumor immunotherapy. By longitudinally monitoring multiple immune cell subsets in peripheral blood and tumor tissue at multiple time points, a spatiotemporal dynamic mathematical model is established. Based on this model, a quantifiable scoring index is constructed to predict the early response / resistance outcome of anti-PD-1 immune checkpoint inhibitor therapy. This method and prediction system can be used to evaluate the efficacy of immunotherapy in animal models and human cohorts. This addresses the technical problems in existing technologies where biomarkers used for predicting immunotherapy efficacy suffer from poor early prediction results and low accuracy due to the static, single-time-point, single-spatial-dimensional evaluation method, insufficient cell coverage, and poor model standardization and cross-cohort transferability.

[0006] The technical solution adopted in this invention is a spatiotemporal immune scoring method, comprising the following steps: An original four-dimensional data matrix is ​​constructed based on individual ID, immune cell type, tissue spatial origin, and sampling time point. The original four-dimensional data matrix contains multiple four-dimensional tensors. Based on the original four-dimensional data matrix, a time dynamic model is constructed for each four-dimensional data individual number-immune cell type combination to obtain time series trend parameters; Based on the four-dimensional tensor at the start and end of the treatment sampling time points, the relative change rate of each immune cell type during the treatment period was calculated using a time dynamic model to obtain the cumulative change parameters. For each immune cell type, the standardized effect size between the treatment-sensitive group and the treatment-resistant group was calculated at each sampling time point to obtain the effect size parameter; Standardized multidimensional weighting coefficients are calculated based on time-series trend parameters, cumulative change parameters, and effect size parameters. Based on standardized multi-dimensional weighting coefficients and a time dynamic model, a spatiotemporal joint score is calculated.

[0007] In some alternative implementations, the immune cell types include CD8+ T cells, myeloid cells, dendritic cells, and B cells.

[0008] In some alternative implementations, the tissue space source includes peripheral blood, tumor tissue, tumor draining lymph nodes, ascites, and cerebrospinal fluid.

[0009] In some optional implementations, the time dynamic model is calculated as follows: p ij (t) = α ij + β ij × t + ε ij (t) Where, α ij The intercept term represents the baseline level of immune cell types; βij The slope term reflects the rate of change in the proportion of immune cells over treatment time and is a core dynamic parameter; ε ij (t) represents the random error term that follows a normal distribution; t is the sampling time point; The time series trend parameters are estimated using the least squares method based on the time dynamic model.

[0010] In some optional implementations, a time-dynamic model is used to calculate the relative rate of change of each immune cell type during treatment using the following formula to obtain the cumulative change parameter: Δrel ij = [p ij (24) - p ij (9)] / p ij (9) × 100% The numbers 24 and 9 represent the number of calendar days.

[0011] In some optional implementations, the standardized effect size between the treatment-sensitive group and the treatment-resistant group is calculated for each immune cell type at each sampling time point using the following formula to obtain the effect size parameter: d j (t) = [μSE j (t) - μAR j (t)] / σpooled j (t) Where μ represents the mean, and the pooled standard deviation is defined as σpooled j (t) = {[(σSE j (t)) 2 + (σAR j (t)) 2 ] / 2} 1 / 2 σ represents the standard deviation.

[0012] In some optional implementations, standardized multidimensional weighting coefficients are calculated based on time-series trend parameters, cumulative change parameters, and effect size parameters, including: Calculate the time series trend weights based on the time series trend parameters; Calculate the cumulative change weight based on the cumulative change parameters; Calculate the effect size weights based on the effect size parameters; Based on the time-series trend weight, cumulative change weight, and effect size weight, multi-dimensional weight coefficients are calculated through sensitivity analysis. The multidimensional weight coefficients are standardized by z-scores in each spatial region to obtain standardized multidimensional weight coefficients.

[0013] In some alternative implementations, a spatiotemporal joint score is calculated based on standardized multidimensional weighting coefficients and a time dynamic model, including: For each individual, a weighted spatial score is first calculated based on standardized multi-dimensional weighting coefficients and a time dynamic model. Then, the weighted spatial score is standardized by z-score within each time point-organizational combination; The standardized scores from various organizational spatial sources are then weighted and fused in the optimal proportion to obtain the final spatiotemporal immunity score.

[0014] Secondly, an application of the spatiotemporal immune scoring method as described in the first aspect in predicting the efficacy of antitumor immunotherapy is provided.

[0015] As can be seen from the above technical solution, the beneficial technical effects of the present invention are as follows: 1. It realizes a paradigm shift from "static single-time-point assessment" to "multi-time-point dynamic tracking", which can obtain more accurate predictive efficacy in the early stage of treatment (such as before and after the second dose after the first dose), avoid ineffective treatment, and save medical resources.

[0016] 2. It can integrate immune dynamic information from multiple spatial sources, taking into account both systemic immune status and local microenvironment status, and make up for the limitations of single-site assessment. Compared with traditional single markers such as PD-L1 expression, T cell exhaustion score, and tumor mutation burden, the spatiotemporal joint score constructed in this invention shows higher and more stable predictive efficacy (AUC) at multiple time points.

[0017] 3. By integrating weights from three dimensions—time-series trend, cumulative change, and effect size—and conducting sensitivity analysis, a robust weight combination is determined, enabling the model to exhibit good robustness to parameter selection and overcoming the problem of excessive dependence on specific fitting parameters in traditional methods. The weight system can be directly transferred to independent validation cohorts and cross-species datasets without refitting, and its predictive performance is highly consistent with that of the training cohort, demonstrating good cross-cohort and cross-species generalization capabilities. Attached Figure Description

[0018] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0019] Figure 1 This is a schematic diagram of the spatiotemporal immune scoring method in the embodiment; Figure 2 This is a schematic diagram of the mouse MC38 colorectal cancer subcutaneous xenograft model process in the examples; Figure 3 The following are tumor growth curves for the AR group, SE group, and control group in Experiment 1 of the example; Figure 4 This is a comparison image of the AR and SE groups showing typical cell type changes in tumor tissue from Experiment 1 in the examples. Figure 5 This is a comparison diagram of peripheral blood changes in some typical cell types in the AR and SE groups in Experiment 1 of the examples; Figure 6 This is a comparison chart of the changes in the proportion of immune cells in the AR group and the SE group during 9-24 days of Experiment 1 in the example. Figure 7 This is a standardized multidimensional weighted comparison chart of various types of immune cells in Experiment 1 of the examples; Figure 8 This is a comparison chart of the STIS scores of the AR group and the SE group in Experiment 1 of the example; Figure 9 This is a comparison chart of the ROC curves for the AR and SE groups in Experiment 1 of the examples; Figure 10 This is a graph showing the prediction accuracy and AUC change of Experiment 1 in the examples; Figure 11 This is a comparison chart of the STIS scores of the AR and SE groups in Experiment 2 of the examples; Figure 12 This is a comparison chart of the ROC curves for the AR and SE groups in Experiment 2 of the examples; Figure 13 This is a graph showing the prediction accuracy and AUC change in Experiment 2 of the examples; Figure 14 This is a comparison chart of tumor tissue treatment response and non-response in Experiment 3 of the examples; Figure 15 This is a comparison chart of peripheral blood treatment response and non-response in Experiment 3 of the examples; Figure 16 This is a comparison chart of the average percentage of immune cells in tumor tissue and peripheral blood in Experiment 3 of the examples. Figure 17 This is a comparison chart of STIS scores for the treatment response groups in Experiment 3 of the examples; Figure 18 The example shows the prediction accuracy curve and ROC curve for Experiment 3. Figure 19 This is a graph showing the predictive efficacy data of other biomarkers (PD-L1, T cell exhaustion score, tumor mutation burden) in Experiment 3 of the examples. Detailed Implementation

[0020] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0021] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0022] Example This embodiment provides a spatiotemporal immune scoring method, including the following steps: Step 1: Construct an original four-dimensional data matrix based on individual ID, immune cell type, tissue spatial origin, and sampling time point. The original four-dimensional data matrix contains multiple four-dimensional tensors. The longitudinally collected immune cell proportion data were organized into a four-dimensional tensor D. i, j, k, t In this context, i represents the individual (subject / animal) ID, j represents the immune cell type (e.g., CD8+ T cells, myeloid cells, dendritic cells, B cells, etc.), k represents the tissue spatial source (e.g., peripheral blood or tumor tissue), and t represents the sampling time point. Those skilled in the art should understand that the tissue spatial source essentially represents the spatial dimension of the data, and is not limited to peripheral blood and tumor tissue sites, but can also be extended to other obtainable body fluids or tissue sites such as tumor draining lymph nodes, ascites, and cerebrospinal fluid.

[0023] The data structure constructed in this step provides a unified mathematical foundation for subsequent time-dynamic modeling and cross-part integration.

[0024] Step 2: Based on the original four-dimensional data matrix, construct a time-dynamic model for each four-dimensional data point combination of individual ID and immune cell type to obtain time-series trend parameters. For each individual ID-immune cell type combination p ij A first-order linear dynamic model was used to fit the change of its proportion over time. The fitting formula is as follows: p ij (t) = α ij + β ij × t + ε ij (t) Where, α ij The intercept term represents the baseline level of immune cell types; β ij The slope term reflects the rate of change in the proportion of immune cells over treatment time and is a core dynamic parameter; ε ij (t) represents the random error term that follows a normal distribution; t is the sampling time point.

[0025] The parameter β of the time dynamic modelij The least squares method is used for estimation: β ij = Cov(t, p ij ) / Var(t), α ij = mean(p ij ) - β ij × mean(t) Where Cov() represents covariance, Var() represents variance, and mean() represents mean.

[0026] In some embodiments, when cell types exhibit significant nonlinear dynamic changes or in application scenarios where longitudinal sampling time points are relatively dense, the first-order linear fitting model in this step can be replaced with a nonlinear dynamic model for fitting, such as spline regression or mixed-effects nonlinear model.

[0027] This step transforms the original multi-timepoint static observations into quantifiable and comparable dynamic trend parameters (slopes), which is the core modeling step in realizing the paradigm shift "from static snapshots to dynamic tracking". pij(t) represents the immune cell types of different individuals, and the pattern of change of the proportion of each type of immune cell in the total number of immune cells over time. β is defined as the time-series trend parameter.

[0028] Step 3: Based on the four-dimensional tensor at the start and end of treatment sampling time points, use a time dynamic model to calculate the relative rate of change of each immune cell type during treatment, and obtain the cumulative change parameters. Four-dimensional tensor D based on the start and end times of treatment i, j, k, t Calculate the absolute change Δabs of each immune cell type during treatment using the following formula. ij With relative change rate Δrel ij : Δabs ij = p ij (24) - p ij (9) Δrel ij = [p ij (24) - p ij (9)] / p ij (9) × 100% The numbers 24 and 9 represent the number of calendar days.

[0029] For the calculation results, when Δ is positive, it indicates that the immune cell type shows an increasing trend during treatment; when Δ is negative, it indicates that the immune cell type shows a contracting trend during treatment; |Δ| reflects the magnitude of the change.

[0030] This step of the calculation captures the net magnitude of immune cell changes throughout the treatment process, Δrel. ij Defined as a cumulative change parameter.

[0031] Step 4: Calculate the standardized effect size between the treatment-sensitive group and the treatment-resistant group for each immune cell type at each sampling time point to obtain the effect size parameter. The training set was constructed as follows: In the C57BL / 6J mouse MC38 colorectal cancer subcutaneous xenograft model, peripheral blood and tumor tissue samples were simultaneously collected from the training cohort (n ≥ 15) of mice receiving anti-PD-1 monoclonal antibody therapy at multiple sampling time points before and during treatment (e.g., 6 sampling time points before and after each administration). The proportions of 14 immune cell subsets were quantified by multicolor flow cytometry.

[0032] Based on the training set data, the standardized effect size d between the SE group (treatment-sensitive group) and the AR group (treatment-resistant group) was calculated at each sampling time point for each immune cell type. j (t): d j (t) = [μSE j (t) - μAR j (t)] / σpooled j (t) Where μ represents the mean, and the pooled standard deviation is defined as σpooled j (t) = {[(σSE j (t)) 2 + (σAR j (t)) 2 ] / 2} 1 / 2 σ represents the standard deviation.

[0033] d j Defined as the effect size parameter.

[0034] In some embodiments, an independent samples t-test can be performed at each time point, and the Benjamini-Hochberg method can be used to correct for false discovery rate (FDR) of the multiple comparison results to obtain a statistically significant set of differentially expressed cell types.

[0035] Step 5: Calculate the standardized multidimensional weighting coefficients based on the time-series trend parameters, cumulative change parameters, and effect size parameters. This step extracts discriminative features for each immune cell type from three complementary dimensions and integrates them into a unified multi-dimensional weighting coefficient. This avoids the information loss caused by relying solely on a single static indicator. The multi-dimensional weighting coefficient includes time-series trend weight, cumulative change weight, and effect size weight; where: The time-series trend weights reflect the differences in the direction and rate of dynamic changes in the proportion of immune cells between the SE and AR groups: Wtrend j = (βSE j - βAR j ) / (|βSE j | + |βAR j | + ε) In the above formula, β is the time series trend parameter, and ε represents the random error.

[0036] Cumulative change weighting reflects the net difference in cell abundance between the two groups at the end of treatment: Wchange j = (ΔSE j - ΔAR j ) / (|ΔSE j | + |ΔAR j | + ε) In the above formula, Δ is the cumulative variation parameter Δrel. ij The calculation is performed, where ε represents the random error.

[0037] Effect size weights are calculated by mapping between-group effect sizes to a bounded interval (-1, 1) using a hyperbolic tangent function, thus preserving the directionality of differences while suppressing the influence of extreme values. In the above formula, Weffect j = tanh(max|d j | / 2) In the above formula, d j This refers to the effect size parameter.

[0038] The multidimensional weight coefficients, incorporating the weights of the three dimensions mentioned above, were calculated and determined through sensitivity analysis. Multiple weight combinations within the range of 0.1-0.5 were evaluated. The results showed extremely low correlation between weight allocation and predictive efficacy (AUC), and the predictive performance remained highly stable within the tested range, indicating good robustness of the model to weight selection. The final comprehensive weights were taken from the center value of this stable interval, with a slightly higher weight assigned to the time-series trend dimension to reflect its biological importance. The multidimensional weight coefficients were calculated using the Wraw... j The calculation method is as follows: Wraw j = 0.4 × Wtrend j + 0.3 × Wchange j + 0.3 × Weffect j In some embodiments, to eliminate differences in weight dimensions between different tissue sites, the multidimensional weight coefficients can be standardized using z-scores within each spatial site to obtain standardized multidimensional weight coefficients Wnorm. jk : Wnorm jk = (Wraw jk - μ k ) / σ k Where, μ k Wraw represents the raw composite weights of all immune cell types within the same spatial location k. jk The arithmetic mean, σ k This indicates the degree of dispersion of the standard deviation mean μk.

[0039] In other embodiments, based on standardized multidimensional weighting coefficients, each cell type can be divided into five functional categories: strong positive correlation, moderate positive correlation, neutral, moderate negative correlation, and strong negative correlation. The classification threshold can be defined as |W_norm|>1.5 for strong, 0.5-1.5 for moderate, and the rest for neutral. This classification step is not necessary for score calculation and is an optional explanatory output.

[0040] Step Six: Calculate the spatiotemporal joint score based on standardized multi-dimensional weighting coefficients and a time dynamic model. For each individual, based on standardized multidimensional weighting coefficients and a time-dynamic model, the weighted spatial score Srawᵢ is first calculated using the following formula. kt : Srawᵢ kt = Σ j Wnorm jk × p ij (t)] / Σ j Wnorm jk | In the above formula, Σ j Wnorm jk This is used to achieve adaptive normalization, making the contributions of positive and negative weights to the final score comparable.

[0041] Then, within each time point-organizational combination, the weighted spatial score is standardized using z-scores to eliminate baseline drift between different time points: Zᵢ kt = (Srawᵢ kt - μ kt ) / σ kt μ ktσ represents the mean of the weighted spatial scores of all individuals within a specific time point (t) and a specific organization (k); kt It represents the standard deviation of the weighted spatial scores of all individuals within a specific time point (t) and a specific organization (k).

[0042] Based on this, the optimal combination was determined through systematic testing of the peripheral blood weighting coefficients within the 0-1 range. The standardized scores of peripheral blood and tumor tissue were then weighted and fused according to the optimal ratio to obtain the final spatiotemporal immune score STIS. STISᵢ t = Blood ×Zᵢ kt + Tumor × Zᵢ kt Wherein, Blood is the peripheral blood weighting coefficient (preferably 0.4 in this embodiment), and Tumor is the tumor tissue weighting coefficient (preferably 0.6 in this embodiment). The peripheral blood weighting coefficient and the tumor tissue weighting coefficient can be re-determined according to the specific application scenario using the same systematic search method. The higher the STIS value, the more inclined the individual is to the treatment response phenotype; the lower the value, the more inclined the individual is to the treatment resistance / non-response phenotype. Those skilled in the art should understand that the above STIS calculation formula uses peripheral blood and tumor tissue as examples. Other spatial source tissues can also be incorporated, such as tumor draining lymph nodes, ascites, cerebrospinal fluid, etc. After incorporating the above three spatial source tissues, STIS has five components. The weights corresponding to these five components can be re-determined according to the specific application scenario using the same systematic search method.

[0043] The technical solution provided in this implementation allows the fixed cell type weighting system constructed in the training queue to be directly applied to independent validation queues or cross-species (e.g., human) datasets without any refitting of the weight parameters, thereby evaluating the model's cross-queue robustness and cross-species universality. For application scenarios such as human single-cell data, a cell type interspecies mapping step is also included (e.g., mapping mouse immune cell subsets to human homologous cell subsets based on marker expression), enabling the same weighting system to be reused across species. Specific experimental data will illustrate this below: Experiment 1: Spatiotemporal immune score modeling and training set validation based on mouse MC38 colorectal cancer model In the C57BL / 6J mouse MC38 subcutaneous xenograft model of colorectal cancer, peripheral blood and tumor tissue samples were simultaneously collected from a training cohort (n=15) of mice receiving anti-PD-1 monoclonal antibody therapy at multiple time points before and during treatment (e.g., 6 time points before and after each administration). Figure 2 and Figure 3 As shown.

[0044] The proportions of 14 immune cell subsets were quantified using multicolor flow cytometry. Based on the gating order of the flow cytometry, the selected immune cell populations included total white blood cells (Live CD45). + ), T cells (CD45) + CD3 + B220 - ), B cells (CD45) + CD3 - B220 + Myeloid cells (CD45) + CD11b + ), Dendritic cell (DC, CD45 + CD11c + MHC-IIhi), DC1 (CD45) + CD11c + MHC-IIhi CD8 + CD11b - DC2 (CD45) + CD11c + MHC-IIhi CD8 - CD11b + After excluding dendritic cells, myeloid cells are further divided into Granulocytes (CD45). + CD11b + Ly6G + Macrophages (CD45) + CD11b + F4 / 80 + ), Ly6Cʰ i -Monocytes (CD45) + CD11b + Ly6Cʰ i ), Ly6Cˡᵒ-Monocytes (CD45 + CD11b + Ly6Clo). Representative cell type variations are shown in [link to relevant documentation]. Figure 4 and Figure 5 The overall trend from day 9 to day 24 is shown in the figure. Figure 6 .

[0045] Those skilled in the art should understand that the selection range of immune cell types can be based on the specific tumor type and is not limited to the 14 cell subpopulations listed in the embodiments of the present invention; the detection platform can also be adjusted and expanded as needed, such as using flow cytometry, single-cell sequencing, mass spectrometry, etc.

[0046] Following the methods described in steps two through six above, the training cohort data is subjected to time-dynamic fitting, cumulative change calculation, and between-group effect size analysis to calculate multi-dimensional weight coefficients, such as... Figure 7 As shown, in peripheral blood, granulocytes acquired the only strong negative weight (W_norm < -1.5), and their dynamic accumulation trend was closely related to the treatment resistance (AR) phenotype, reflecting that the continuous mobilization of myeloid suppressor components in peripheral blood is the core immunological feature of treatment failure. Macrophages and myeloid cells also showed a moderate negative weight overall, suggesting that they were more prominently enriched over time in the AR group. B cells also acquired a moderate negative weight in peripheral blood. Although the proportion of SE cells was higher than that in the AR group at some time points, peripheral blood B cells in the SE group experienced more significant dynamic fluctuations and declines during treatment, while those in the AR group remained relatively stable and showed an increasing trend. This led to B cell changes being more likely to be associated with the AR phenotype when assessing the overall temporal dynamic dimension. This phenomenon suggests that the dynamic trajectory of peripheral blood B cells, rather than the absolute abundance at a certain time point, is the key information dimension for their association with treatment response. In contrast, T cells and CD4 cells... + T cells and total white blood cells (Live CD45) + (Moderate Positive) Ly6Clo monocytes, CD8 + T cells and monocytes also showed positive weights, with high expression tending to predict the treatment-sensitive (SE) phenotype. The weight of DC-related subsets in peripheral blood was close to neutral, suggesting that their contribution to the discrimination of SE / AR in peripheral blood was relatively limited.

[0047] In the tumor microenvironment, the weighting pattern is clearer and the polarization is more pronounced. For example... Figure 7 As shown, CD8 + T cells and total tumor-infiltrating leukocytes (Live CD45) + The highest strong positive weight (W_norm > 1.5) is obtained, which is the most core positive feature for predicting treatment response, and is related to CD8 in the tumor of the SE group. + The dynamic pattern of persistent T cell enrichment closely matches this finding. DCs and DC2 also showed moderate positive weights, suggesting that the functional activation of antigen-presenting cells in the tumor microenvironment participates in the formation of an effective immune response. In contrast, Ly6C hi monocytes, macrophages, myeloid cells, monocytes, and granulocytes all received moderate negative weights; their enrichment in the tumor microenvironment collectively constitutes a marker of the immunosuppressive microenvironment and is closely related to the AR phenotype.

[0048] STIS scores were calculated at each time point, and the area under the receiver operating characteristic (AUC) and classification accuracy were used to systematically evaluate the predictive efficacy of STIS at each time point. Simultaneously, its relative predictive advantage compared to traditional single biomarkers (such as PD-L1 expression, T cell exhaustion score, and TMB) was assessed. Results are as follows: Figure 8 , Figure 9 and Figure 10 As shown: Before treatment (before the first dose), STIS had no discriminative power between the two groups (AUC≈0). However, before and after the second dose following the first dose, the discriminative power of STIS reached a medium-high level (AUC above 0.8), and further improved and remained at a very high level at subsequent time points (AUC close to or reaching 1.0), significantly earlier than the time point when response / resistance was visually distinguished based on phenotypic indicators such as tumor volume.

[0049] Experiment 2: Application of cross-queue weight transfer in independent verification queues The fixed weight coefficients constructed in the training cohort of Experiment 1 were directly applied without any adjustment to a new independent mouse validation cohort (n=15). STIS scores at each time point in this cohort were calculated, and its predictive efficacy was evaluated. Results are as follows: Figure 11 , Figure 12 and Figure 13 As shown, the STIS trajectory change pattern, inter-group separation time window, and prediction efficacy (AUC) of the validation queue are highly consistent with those of the training queue, confirming that the standardized multi-dimensional weight coefficient and spatiotemporal immune scoring method constructed in this invention has good cross-cohort transferability and robustness.

[0050] Experiment 3: Weight Transfer and Clinical Translation Applications of Cross-Species (Human) Single-Cell Datasets The immune cell weight coefficients established in the training cohort were transferred to the human colorectal cancer single-cell sequencing dataset (GSE236581, including 22 patients, 19 responders and 3 non-responders) through interspecies homology mapping of cell types (e.g., mouse DC1 / DC2 corresponding to human cDC1 / cDC2, mouse monocyte subsets corresponding to human classical / non-classical monocytes, etc.). STIS scores were calculated based on peripheral blood and tumor tissue data from two available time points: before treatment and early treatment. Results are as follows: Figures 14-19 As shown, STIS significantly distinguished between the responding and non-responding groups before and in the early stages of treatment (AUC approximately 0.91 and 0.84, respectively), and its predictive efficacy was significantly better than that of traditional single biomarkers such as tumor PD-L1 expression, T cell exhaustion score, and tumor mutational burden at the same cohort and time point, thus verifying the cross-species clinical translation potential of the technical solution of this invention.

[0051] By adopting the technical solution of this application, a paradigm shift from "static single-time-point assessment" to "multi-time-point dynamic tracking" has been achieved, which can obtain more accurate predictive efficacy in the early stage of treatment (such as before and after the second dose after the first dose), avoid ineffective treatment, and save medical resources.

[0052] It can integrate immune dynamic information from multiple spatial sources, taking into account both systemic immune status and local microenvironment status, and make up for the limitations of single-site assessment. Compared with traditional single markers such as PD-L1 expression, T cell exhaustion score, and tumor mutation burden, the spatiotemporal joint score constructed in this invention shows higher and more stable predictive efficacy (AUC) at multiple time points, which is significantly better than the comparison method.

[0053] The robust weight combination determined through the integration of weights based on time-series trends, cumulative changes, and effect sizes, along with sensitivity analysis, makes the model robust to parameter selection, overcoming the problem of excessive dependence on specific fitting parameters in traditional methods. The weight system can be directly transferred to independent validation cohorts and cross-species datasets without refitting, and the predictive performance is highly consistent with the training cohort, demonstrating good cross-cohort and cross-species generalization capabilities.

[0054] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A spatiotemporal immune scoring method, characterized in that, Includes the following steps: An original four-dimensional data matrix is ​​constructed based on individual ID, immune cell type, tissue spatial origin, and sampling time point. The original four-dimensional data matrix contains multiple four-dimensional tensors. Based on the original four-dimensional data matrix, a time dynamic model is constructed for each four-dimensional data individual number-immune cell type combination to obtain time series trend parameters; Based on the four-dimensional tensor at the start and end of the treatment sampling time points, the relative change rate of each immune cell type during the treatment period was calculated using a time dynamic model to obtain the cumulative change parameters. For each immune cell type, the standardized effect size between the treatment-sensitive group and the treatment-resistant group was calculated at each sampling time point to obtain the effect size parameter; Standardized multidimensional weighting coefficients are calculated based on time-series trend parameters, cumulative change parameters, and effect size parameters. Based on standardized multi-dimensional weighting coefficients and a time dynamic model, a spatiotemporal joint score is calculated.

2. The spatiotemporal immune scoring method according to claim 1, characterized in that, The immune cell types include CD8+ T cells, myeloid cells, dendritic cells, and B cells.

3. The spatiotemporal immune scoring method according to claim 1, characterized in that, The tissue sources include peripheral blood, tumor tissue, tumor draining lymph nodes, ascites, and cerebrospinal fluid.

4. The spatiotemporal immune scoring method according to claim 1, characterized in that, The calculation formula for the time dynamic model is as follows: p ij (t) = a ij + b ij × t + e ij (t) Where, α ij The intercept term represents the baseline level of immune cell types; β ij The slope term reflects the rate of change in the proportion of immune cells over treatment time and is a core dynamic parameter; ε ij (t) represents the random error term that follows a normal distribution; t is the sampling time point; The time series trend parameters are estimated using the least squares method based on the time dynamic model.

5. The spatiotemporal immune scoring method according to claim 4, characterized in that, The relative rate of change of each immune cell type during treatment was calculated using the time-dynamic model, and the cumulative change parameter was obtained by the following formula: Δrel ij = [p ij (24) - p ij (9)] / p ij (9) × 100% The numbers 24 and 9 represent the number of calendar days.

6. The spatiotemporal immune scoring method according to claim 1, characterized in that, The standardized effect size between the treatment-sensitive group and the treatment-resistant group was calculated for each immune cell type at each sampling time point using the following formula to obtain the effect size parameter: d j (t) = [μSE j (t) - μAR j (t)] / σpooled j (t) Where μ represents the mean, and the pooled standard deviation is defined as σpooled j (t) = {[(σSE j (t)) 2 + (σAR j (t)) 2 ] / 2} 1 / 2 σ represents the standard deviation.

7. The spatiotemporal immune scoring method according to claim 1, characterized in that, Standardized multidimensional weighting coefficients are calculated based on time-series trend parameters, cumulative change parameters, and effect size parameters, including: Calculate the time series trend weights based on the time series trend parameters; Calculate the cumulative change weight based on the cumulative change parameters; Calculate the effect size weights based on the effect size parameters; Based on the time-series trend weight, cumulative change weight, and effect size weight, multi-dimensional weight coefficients are calculated through sensitivity analysis. The multidimensional weight coefficients are standardized by z-scores in each spatial region to obtain standardized multidimensional weight coefficients.

8. The spatiotemporal immune scoring method according to claim 1, characterized in that, Based on standardized multi-dimensional weighting coefficients and a time-dynamic model, a spatiotemporal joint score is calculated, including: For each individual, a weighted spatial score is first calculated based on standardized multi-dimensional weighting coefficients and a time dynamic model. Then, the weighted spatial score is standardized by z-score within each time point-organizational combination; The standardized scores from various organizational spatial sources are then weighted and fused in the optimal proportion to obtain the final spatiotemporal immunity score.

9. The spatiotemporal immune scoring method as described in claim 1, characterized in that, Application in predicting the efficacy of antitumor immunotherapy.

Images