Tumor early screening and typing early warning system based on multi-omics data association analysis

Abstract

The invention relates to the technical field of bioinformatics and clinical medicine, and discloses a multi-omics data association analysis-based tumor early screening and typing early warning system, which comprises a data acquisition and preprocessing module for integrating standardized longitudinal multi-omics data; the dynamics and topology analysis module is used for generating topology fingerprints representing dynamic behaviors of the system through state space reconstruction and persistent coherence analysis; the causal inference and risk assessment module is used for calculating critical moderation indexes in parallel to synthesize risk indexes and constructing a dynamic causal network; and a collaborative diagnosis and report generation module. According to the system, risk indexes derived by critical moderation, topological fingerprints and a dynamic causal network are creatively combined, multi-modal information fusion is carried out through a collaborative diagnosis unit, and finally a comprehensive early warning report is generated. According to the invention, the accuracy and reliability of early risk early warning of tumors can be obviously improved, and a mechanism-level traceability basis is provided for clinical intervention.

Description

Technical Field

[0001] This invention relates to the fields of bioinformatics and clinical medicine, specifically to an early tumor screening and typing warning system based on multi-omics data association analysis. Background Technology

[0002] Cancer is a major disease that seriously threatens human health, and its early detection and diagnosis are crucial for improving patient survival rates and prognosis. With the development of technologies such as high-throughput sequencing, it has become possible to explore the mechanisms of cancer development and progression at the systemic level by integrating multi-dimensional omics data such as genomics, transcriptomics, proteomics, and metabolomics. This brings new opportunities for ultra-early cancer screening and early warning.

[0003] However, current tumor screening methods based on multi-omics data still have many limitations. Existing mainstream methods typically rely on differential analyses of healthy and patient populations at specific time points to identify static biomarkers. This cross-sectional research approach is inherently static and struggles to capture the intrinsic patterns of tumors as a dynamic evolutionary process. While it may effectively identify established lesions, it is insufficiently sensitive to the gradual transition of the system from health to instability before disease onset, thus missing the optimal window for early warning and intervention.

[0004] Even though some studies have begun to utilize longitudinal data, their analytical methods are often limited to traditional statistical models. These models typically assume that system changes are linear, making it difficult to effectively reveal the complex nonlinear dynamic characteristics exhibited by complex biological systems when they approach the "critical point" of disease, such as the "critical slowing" phenomenon caused by weakened system resilience.

[0005] Furthermore, existing technologies for uncovering intermolecular interactions mostly focus on discovering molecular features associated with disease states. However, these correlation analyses often fail to effectively distinguish between causal relationships and concomitant effects. This results in the inclusion of a large number of non-core, passively changing molecules in the analysis results, making it difficult to pinpoint the key regulatory nodes and pathways that truly drive disease progression, thus limiting the development of subsequent precision intervention strategies.

[0006] More importantly, existing technologies often separate the analytical dimensions of risk quantification, state change identification, and intrinsic mechanism inference, lacking a unified framework that can organically integrate these multimodal features that originate from different theories and contain different information. Therefore, it is impossible to form a comprehensive diagnostic conclusion that can both accurately predict risks and clearly trace the source mechanism. Summary of the Invention

[0007] The technical problem to be solved by this invention is that existing multi-omics data association analysis methods, when applied to early cancer screening, have technical defects such as insufficient analysis of the dynamic evolution process of the disease, difficulty in distinguishing between causal driving relationships and statistical correlations between molecules, and warning signals are usually only identified after clinical symptoms appear.

[0008] To address the aforementioned technical problems, the first aspect of this invention provides a tumor early screening and typing warning system based on multi-omics data association analysis.

[0009] The system includes:

[0010] The data acquisition and preprocessing module acquires and processes longitudinal multi-omics data from an individual.

[0011] The dynamics and topology analysis module generates an evolutionary trajectory point cloud for the individual based on the longitudinal multi-omics data, and generates a topological fingerprint based on the evolutionary trajectory point cloud;

[0012] The causal inference and risk assessment module infers a dynamic causal network based on the longitudinal multi-omics data and calculates a critical risk index.

[0013] The collaborative diagnosis and report generation module integrates the topological fingerprint, the dynamic causal network, and the critical risk index to generate a comprehensive early warning report.

[0014] In an optional embodiment, the dynamics and topology analysis module includes a state space reconstruction unit, a reconstruction parameter optimization unit, and a topology fingerprint generation unit. The state space reconstruction unit reconstructs the evolutionary trajectory point cloud from one or more time series data points in the longitudinal multi-omics data using a time-delay embedding method.

[0015] Specifically, for a certain time series x of an individual k k (t), at time point t j The system state vector S k (t j The following formula is given:

[0016] S k (t j )=[x k (t j ),x k (t j +τ k ),x k (t j +2τ k ),…,x k (t j +(m k-1)τ k )];

[0017] Where, m k For the embedding dimension; τ k S is the time delay parameter. k (t j ) represents individual k at time point t j The state vector, which is m k A point in 3D space; k represents the index number of the individual; t j Let x represent the j-th discrete observation time point; k (t) represents one or more base time series selected from the longitudinal multi-omics data of individual k. The reconstruction parameter optimization unit determines the embedding dimension m for the state space reconstruction unit before performing time-delay embedding. k With the time delay parameter τ k .

[0018] In one implementation, the reconstruction parameter optimization unit determines the time delay parameter τ using the average mutual information method. k The embedding dimension parameter m is determined using a pseudo-nearest neighbor algorithm. k The topological fingerprint generation unit analyzes the evolutionary trajectory point cloud to generate the topological fingerprint.

[0019] In one implementation, the topological fingerprint generation unit applies a persistent cohomology algorithm to analyze the evolutionary trajectory point cloud, quantizing the lifecycle of the topological features of the evolutionary trajectory point cloud into a persistent graph or persistent barcode, which serves as the topological fingerprint.

[0020] In an optional embodiment, the causal inference and risk assessment module includes a dynamic causal network construction unit, a critical slowing-down calculation unit, and a risk index synthesis unit. The dynamic causal network construction unit analyzes the time-series relationships between different molecular features in the longitudinal multi-omics data to infer the dynamic causal network.

[0021] In one implementation, the dynamic causal network construction unit applies a convergent cross-mapping algorithm to infer the dynamic causal network. The critical slowing computation unit calculates one or more statistical indicators of the longitudinal multi-omics data within a sliding time window.

[0022] In one implementation, the critical slowing calculation unit calculates the first-order autocorrelation coefficient and variance of the longitudinal multi-omics data within the sliding time window as one or more statistical indicators, and determines that a synchronous and continuous upward trend of the first-order autocorrelation coefficient and variance is a warning signal of system instability. Wherein, at time point t... jWithin a window of size W at the end, the first-order autocorrelation coefficient AC1(t) j The formula for calculating ) is:

[0023]

[0024] Among them, AC1(t) j ) indicates at time point t j The first-order autocorrelation coefficient calculated within the sliding time window ending at the end; j represents the index of the end time point of the current sliding time window; W represents the size of the sliding time window, i.e., the number of consecutive time points included in the calculation of the statistical indicator; i represents the time point index when performing the summation operation within the sliding time window; pc k (t i ) represents individual k at time point t i The principal component values ​​of the state vector, or other one-dimensional time series values ​​that can represent the system state; This indicates that at time point t j Within the sliding time window ending, all pcs k (t i The arithmetic mean of the values; k represents the index number of the individual;

[0025] Variance Var(t) j The formula for calculating ) is:

[0026]

[0027] Among them, pc k (t i For individual k at time point t i Principal component values ​​of the state vector Var(t) represents the mean value within that time window. j ) indicates at time point t j The variance is calculated within a sliding time window ending at the end. The risk index synthesis unit synthesizes the critical risk index based on the changing trends of the one or more statistical indicators.

[0028] In an optional embodiment, the collaborative diagnosis and report generation module includes a multimodal information fusion unit, a risk tracing and classification unit, and a visualization report generation unit. The multimodal information fusion unit performs collaborative diagnosis on the critical risk index, the topological fingerprint, and the dynamic causal network.

[0029] In one implementation, the multimodal information fusion unit uses the critical risk index as a primary criterion for risk level determination; uses changes in topological features in the topological fingerprint as a secondary criterion for confirming system stability; and uses the dynamic causal network as a tertiary criterion for risk tracing and classification.

[0030] The risk tracing and classification unit performs risk tracing and potential classification based on the driving nodes and paths in the dynamic causal network.

[0031] The visualization report generation unit generates a comprehensive early warning report that includes risk curves, topology diagrams, and network diagrams.

[0032] This invention provides a tumor early screening and subtyping warning system based on multi-omics data association analysis.

[0033] It has the following beneficial effects:

[0034] 1. This invention reconstructs the longitudinal multi-omics data of individuals into an evolutionary trajectory point cloud and calculates its critical slowing index, which can capture early signals of the decline in stability of a healthy system before it transitions to a disease state. This judgment method based on system dynamic stability can issue an early warning before traditional biomarker indicators reach abnormal thresholds, thereby significantly advancing the time window for risk identification.

[0035] 2. This invention adopts a multimodal collaborative diagnostic mechanism, which integrates and analyzes the quantified critical risk index, the topological fingerprint describing the trajectory pattern, and the dynamic causal network revealing the intrinsic driving relationship. By cross-validating information from different analytical dimensions, it effectively avoids the one-sidedness and randomness that may be caused by a single indicator, thereby reducing the probability of misjudgment and enhancing the confidence of the final warning report.

[0036] 3. By constructing a dynamic causal network, this invention can not only determine whether a risk exists, but also identify upstream causal drivers that cause system instability. This risk tracing capability provides clear biological clues for understanding individual pathogenesis. The identified driving nodes can serve as potential targets for subsequent precision medicine verification or personalized prevention and intervention. Attached Figure Description

[0037] Figure 1 This is a structural block diagram of a tumor early screening and subtyping warning system based on multi-omics data association analysis, according to an embodiment of the present invention.

[0038] Figure 2 This is a structural block diagram of a data acquisition and preprocessing module according to an embodiment of the present invention;

[0039] Figure 3 This is a structural block diagram of the dynamics and topology analysis module according to an embodiment of the present invention;

[0040] Figure 4 This is a structural block diagram of a causal inference and risk assessment module according to an embodiment of the present invention;

[0041] Figure 5 This is a structural block diagram of a collaborative diagnosis and report generation module according to an embodiment of the present invention. Detailed Implementation

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] See attached document Figure 1 , Figure 1 This is a structural block diagram of a tumor early screening and subtyping warning system based on multi-omics data association analysis according to an embodiment of the present invention. The system provided by the present invention may include: a data acquisition and preprocessing module 100, a dynamics and topology analysis module 200, a causal inference and risk assessment module 300, and a collaborative diagnosis and report generation module 400.

[0044] In one specific embodiment, the data acquisition and preprocessing module 100 acquires longitudinal multi-omics data of an individual at multiple discrete time points. The longitudinal multi-omics data is preprocessed by this module, including standardization and quality control, to generate standardized multivariate time series data, which is then output to the dynamics and topology analysis module 200 and the causal inference and risk assessment module 300.

[0045] The dynamics and topology analysis module 200 receives standardized multivariate time series data from the data acquisition and preprocessing module 100. This module first reconstructs the time series data into an evolutionary trajectory point cloud of an individual in the state space using a time delay embedding method.

[0046] Specifically, at time point t j The system state vector S k (t j The following formula is given:

[0047] S k (t j )=[x k (t j ),x k (t j +τ k ),x k (t j +2τ k ),…,x k (t j +(m k -1)τ k )];

[0048] Where, m k For the embedding dimension; τ k S is the time delay parameter. k (t j ) represents individual k at time point t j The state vector, which is m k A point in 3D space; k represents the index number of the individual; t j Let x represent the j-th discrete observation time point; k (t) represents one or more base time series selected from the longitudinal multi-omics data of individual k.

[0049] Subsequently, the dynamics and topology analysis module 200 applies continuous coherence analysis to the generated evolutionary trajectory point cloud to quantify its topological structure and generate a topological fingerprint. This topological fingerprint is then output to the collaborative diagnosis and report generation module 400.

[0050] The causal inference and risk assessment module 300 also receives standardized multivariate time series data from the data acquisition and preprocessing module 100. This module performs two parallel computational tasks. First, it infers the causal relationships between molecular features in the multivariate time series data, constructing a dynamic causal network. Second, within a sliding time window, it calculates the critical slowing index of the time series data, including the first-order autocorrelation coefficient and variance. The first-order autocorrelation coefficient AC1(t) is... j The formula for calculating ) is:

[0051]

[0052] Among them, AC1(t) j ) indicates at time point t j The first-order autocorrelation coefficient calculated within the sliding time window ending at the end; j represents the index of the end time point of the current sliding time window; W represents the size of the sliding time window, i.e., the number of consecutive time points included in the calculation of the statistical indicator; i represents the time point index when performing the summation operation within the sliding time window; pc k (t i ) represents individual k at time point t i The principal component values ​​of the state vector, or other one-dimensional time series values ​​that can represent the system state; This indicates that at time point t j Within the sliding time window ending, all pcs k (t i The arithmetic mean of the values; k represents the index number of the individual;

[0053] Variance Var(t) j The formula for calculating ) is:

[0054]

[0055] Among them, pc k (t i For individual k at time point t i Principal component values ​​of the state vector Var(t) represents the mean value within that time window. j ) indicates at time point t j The variance is calculated within the sliding time window at the end. This module synthesizes a critical risk index based on the changing trends of the first-order autocorrelation coefficient and variance. Both the dynamic causal network and the critical risk index are output to the collaborative diagnosis and report generation module 400.

[0056] The collaborative diagnosis and report generation module 400 receives the analysis results from the aforementioned modules, namely topological fingerprints, dynamic causal networks, and critical risk indices. This module integrates these three types of heterogeneous information, performs collaborative diagnostic analysis, and ultimately generates a comprehensive early warning report that includes risk level, trajectory morphology analysis, and causal relationship tracing.

[0057] See attached document Figure 2 , Figure 2 This is a structural block diagram of a data acquisition and preprocessing module 100 according to an embodiment of the present invention. The module includes a vertical data interface unit 110, a data standardization unit 120, and a data quality control unit 130.

[0058] The system execution flow of this embodiment of the invention begins with acquiring individual-related biological data through the data acquisition and preprocessing module 100. Specifically, the longitudinal data interface unit 110 is responsible for receiving data for the same individual at multiple different time points (denoted as t1, t2, ..., t...). N , where N is the total number of collections and N≥2) is the detection data generated from periodically collected biological samples.

[0059] Biological samples are obtained through non-invasive or minimally invasive methods; in one specific implementation, the biological sample is a peripheral blood sample. The detection data encompasses multiple omics levels, constituting multi-omics data. Multi-omics data includes, but is not limited to: methylomics data and genomic variation data derived from cell-free DNA (cfDNA) in plasma, plasma proteomics data, and serum metabolomics data.

[0060] For an individual k, at any time point t j The acquired data constitutes a P-dimensional molecular feature vector O. k (t j )={o k,1 (t j ),o k,2 (tj ),...,o k,P (t j )}, where P is the total number of molecular features detected. Therefore, the complete dataset for this individual k is a sequence of feature vectors {O} ordered by time. k (t1),O k (t2),...,O k (t N This sequence is the longitudinal multi-omics data and serves as the basis for subsequent analysis.

[0061] After acquiring the raw longitudinal multi-omics data through the longitudinal data interface unit 110, the data acquisition and preprocessing module 100 performs a series of preprocessing procedures on the data through its data standardization unit 120 and data quality control unit 130 to generate a standard data format that can be used by subsequent analysis modules.

[0062] Data standardization unit 120 standardizes the dimensions of different molecular features. In one specific implementation, data standardization unit 120 calculates a mean μ for each molecular feature P (where p∈{1,…,P}) using its observations across all individuals and all time points. p and a standard deviation σ p Subsequently, the Z-score normalization method was applied to normalize each original observation o. k,p (t j Convert ) to the standardized value o′ k,p (t j The conversion process is defined by the following formula:

[0063]

[0064] Among them, o′ k,p (t j ) is the p-th molecular characteristic of individual k at time point t. j The standardized value; o k,p (t j ) represents its corresponding original observation value; μ p σ is the mean of the p-th molecular feature across the entire dataset; p It is the standard deviation of the p-th molecular feature over the entire dataset.

[0065] The data quality control unit 130 is responsible for handling technical deviations in the data. This unit first imputes missing values ​​in the data.

[0066] In one implementation, the K-Nearest Neighbors (KNN) interpolation algorithm is used. For a missing data point, the system identifies the k other samples in the sample space that have the most similar features to it, and uses the weighted average of these k samples on that feature to fill the missing value.

[0067] Next, the data quality control unit 130 corrects for non-biological variations introduced by different measurement batches. In one embodiment, the data quality control unit 130 applies the ComBat algorithm, using known batch information as covariates, to adjust the data to eliminate systematic bias caused by batch effects while preserving the inherent biological differences in the data.

[0068] After the above standardization, missing value imputation and batch effect correction processes, the data acquisition and preprocessing module 100 outputs a set of standardized, complete and internally consistent longitudinal multi-omics data and transmits it to the dynamics and topology analysis module 200 and the causal inference and risk assessment module 300.

[0069] Please refer to the appendix. Figure 3 , Figure 3 This is a structural block diagram of a dynamics and topology analysis module 200 according to an embodiment of the present invention. The module may include a state space reconstruction unit 210, a reconstruction parameter optimization unit 220, and a topology fingerprint generation unit 230.

[0070] The dynamics and topology analysis module 200 receives standardized longitudinal multi-omics data processed by the data acquisition and preprocessing module 100. The function of this module is to transform high-dimensional, discrete observation data into a low-dimensional continuous state-space trajectory that can reflect the inherent dynamic evolution of the system.

[0071] In one specific embodiment, the state-space reconstruction unit 210 performs this transformation process. First, the unit takes a P-dimensional multivariate time series data {O′} from an individual k. k (t j Extract one or more basic time series x from the )}. k (t).

[0072] In one implementation, the state-space reconstruction unit 210 performs principal component analysis (PCA) on the P-dimensional eigenvectors at each time point and selects the time series of the first principal component as the base time series x. k (t), this sequence captures the main direction of variation in the original data.

[0073] Next, the state space reconstruction unit 210 applies the time delay embedding method, utilizing the base time series xk (t) is used to construct a series of state vectors. At any time point t j , an m k dimensional state vector S k (t j () is defined by the following formula:

[0074] S k (t j )=[x k (t j ),x k (t j +τ k ),x k (t j +2τ k ),...,x k (t j +(m k -1)τ k The formula

[0075] By extracting data from the base time series starting at time t j Start, with a time delay τ k For a continuous m interval k Each data point forms a vector. The embedding dimension m in this process... k and time delay τ k These are key parameters, which are predetermined by the reconstructed parameter optimization unit 220.

[0076] By performing the above calculations at all available time points, the state-space reconstruction unit 210 generates a state-space reconstruction unit {S} for individual k, consisting of a sequence of state vectors {S}. k (t1),S k (t2),...,S k (t N The set consisting of m) k This forms a point cloud in 3D space, which is the evolutionary trajectory point cloud. This evolutionary trajectory point cloud will be transmitted to the topological fingerprint generation unit 230 for subsequent topological structure analysis.

[0077] Before the state space reconstruction unit 210 performs reconstruction, the dynamics and topology analysis module 200, through its reconstruction parameter optimization unit 220, adaptively determines the optimal time delay parameter τ for each volume k. k and embedding dimension parameter m k This process ensures that the subsequently generated evolutionary trajectory point cloud can accurately reproduce the topology of the original dynamical system.

[0078] First, the reconfiguration parameter optimization unit 220 determines the time delay parameter τ. kIn one specific implementation, the reconstructed parameter optimization unit 220 calculates the underlying time series x. k The average mutual information (AMI) of (t) is calculated as follows: The relationship between AMI and time delay τ is:

[0079]

[0080] in:

[0081] I(τ) is the average mutual information value when the time delay is τ;

[0082] τ is a variable representing time delay;

[0083] p(x(t h )) is a time series with value x(t) h The probability density at time ( );

[0084] p(x(t h ),x(t h +τ)) represents the time series at time point t. h and t h The joint probability density of +τ for a specific pair of values;

[0085] x(t h ) represents the basic time series x(t) at time point t. h The observed values.

[0086] h,l represents the summation index used to calculate the probability density, traversing all discretized data intervals.

[0087] This unit determines the value of τ corresponding to the first local minimum point of the calculated I(τ) function curve as the optimal time delay parameter τ. k .

[0088] In determining τ k Subsequently, the reconstruction parameter optimization unit 220 applies the False Nearest Neighbors (FNN) algorithm to determine the optimal embedding dimension m. k The algorithm is executed as follows: For a given test dimension m, the reconstruction parameter optimization unit 220 finds the nearest neighbor for each state vector in the m-dimensional reconstruction space. Then, the dimension is increased to m+1, and the distance between these neighbor pairs is recalculated. If the relative increase in distance between a pair of neighbors after the dimension increase exceeds a preset threshold, the pair of neighbors is determined to be pseudo-near neighbors.

[0089] The parameter optimization unit 220 starts from m=1 and incrementally calculates the proportion of pseudo-near neighbors under different test dimensions. When this proportion first drops below a preset tolerance threshold (e.g., 1%), the corresponding test dimension m is determined as the optimal embedding dimension m. k .

[0090] After completing the above calculations, the reconstructed parameter optimization unit 220 will determine the parameter pairs (τ) k ,m k The data is transmitted to the state space reconstruction unit 210 for state space reconstruction.

[0091] After receiving the evolution trajectory point cloud generated by the state space reconstruction unit 210, the dynamics and topology analysis module 200 performs quantitative analysis on the geometry and topology of the point cloud through its topology fingerprint generation unit 230.

[0092] In one specific implementation, the topological fingerprint generation unit 230 applies a persistent cohomology algorithm. This algorithm first constructs a filtered simplex complex based on the evolutionary trajectory point cloud.

[0093] Specifically, the topological fingerprint generation unit 230 defines a neighborhood sphere with radius ∈ for each point in the point cloud. As the radius ∈ increases continuously from 0, whenever the neighborhood spheres of k+1 points have a common intersection, a k-dimensional simplex is formed between these k+1 points. This process generates a nested sequence of simplex complexes that increases with ∈, which serves as a filter.

[0094] During this filtering process, the topological fingerprint generation unit 230 continuously tracks the appearance and disappearance of topological features. Topological features include 0-dimensional features (connected components), 1-dimensional features (loops or cycles), and higher-dimensional holes. A topological feature has a radius value ∈ birth The first time this feature is formed is denoted as the "birth" of the feature; as the radius value grows to ∈ death When a feature disappears, it is filled by a higher-dimensional simplex and is denoted as the "extinction" of that feature.

[0095] This unit records the lifetime of each topological feature as an interval [∈ birth ,∈ death All these sets of intervals can be visualized and quantified in two equivalent ways: one is persistence barcode, where each interval is represented as a value from ∈ birth Extend to ∈ death One is a bar graph; the other is a persistence graph, where each interval is represented as a point in a two-dimensional coordinate system (∈).birth ,∈ death ).

[0096] The generated persistent graph or persistent barcode can quantitatively describe the topological structure of the evolutionary trajectory point cloud at different scales. This persistent graph or persistent barcode constitutes the topological fingerprint of the individual and is output by the topological fingerprint generation unit 230 to the collaborative diagnosis and report generation module 400 for subsequent system stability assessment.

[0097] Please refer to the appendix. Figure 4 , Figure 4 This is a structural block diagram of a causal inference and risk assessment module 300 according to an embodiment of the present invention. The module may include a dynamic causal network construction unit 310, a critical slowing-down calculation unit 320, and a risk index synthesis unit 330.

[0098] The causal inference and risk assessment module 300 receives standardized longitudinal multi-omics data output from the data acquisition and preprocessing module 100. The data contains time series data of multiple molecular features.

[0099] The function of the dynamic causal network building unit 310 is to analyze the interaction between these molecular feature time series in order to infer a dynamic network that characterizes causal influence.

[0100] In one specific implementation, the dynamic causal network building unit 310 applies the Convergent Cross Mapping (CCM) algorithm to perform this causal inference.

[0101] For any two molecular feature time series, denoted as X(t) and Y(t), the dynamic causal network construction unit 310 performs the following steps to determine whether X has a causal influence on Y. First, the unit uses the time-delay embedding method to reconstruct their respective state-space manifolds from the time series X(t) and Y(t), denoted as M. X and M Y .

[0102] To examine the causal effect of X on Y, this unit is based on manifold M. Y To predict the value of the time series X(t). Specifically, for M Y For each point on the network (i.e., a state vector composed of values ​​in the Y(t) time series), the unit identifies several nearest neighbors within its neighborhood. Then, using the actual values ​​of X(t) corresponding to these neighboring points in time, an estimate of X(t) is calculated through a weighted average method, denoted as […].

[0103] Then, the unit calculates the original time series X(t) and the above-mentioned estimated value series. The Pearson correlation coefficient between them is denoted as ρ. X,Y This correlation coefficient measures the effectiveness of using manifold M. Y The accuracy of predicting X(t).

[0104] This unit further verifies whether this prediction accuracy is convergent. Specifically, this unit increases the amount of material used to construct the manifold M stepwise. Y Given the length of the time series (i.e., the library size L), repeat the above prediction process and calculate the corresponding correlation coefficient ρ. X,Y If the correlation coefficient ρ increases with the increase of the library size L... X,Y If the value of X is significantly greater than zero and tends to a stable saturation value, then it is determined that molecular characteristic X has a causal influence on molecular characteristic Y. Conversely, if it is necessary to test the causal influence of Y on X, then a symmetric calculation process should be performed.

[0105] The dynamic causal network construction unit 310 repeatedly performs the above pairwise causal inference on all preset molecular feature pairs in the data. Finally, the dynamic causal network construction unit 310 integrates all determined causal relationships into a directed weighted graph. In this graph, nodes represent individual molecular features, directed edges represent the causal influence direction between features, and the weights of the edges are determined by the converged correlation coefficient values. This graph is the dynamic causal network, which is then output to the collaborative diagnosis and report generation module 400.

[0106] In parallel with the construction of the dynamic causal network, the critical slowing calculation unit 320 of the causal inference and risk assessment module 300 is responsible for quantitatively assessing the dynamic changes in system stability. This unit receives a one-dimensional time series representing the overall state of the system. In one implementation, this series is the first principal component time series of the multi-omics data output by the data acquisition and preprocessing module 100, denoted as pc. k (t).

[0107] The critical slowing computation unit 320 uses a sliding time window method to analyze the time series pc. k (t). Specifically, the critical slowing calculation unit 320 defines a time window of width W and slides this window along the time axis from the beginning to the end of the sequence. For time point t j For each window with the endpoint, this cell calculates two statistical indicators for the data within the window: the first-order autocorrelation coefficient AC1(t). j ) and variance Var(t j ).

[0108] By repeating this calculation at all time points, the critical slowing calculation unit 320 generates two new time series for individual kk: an autocorrelation coefficient series AC1(t) and a variance series Var(t). These two series are then transmitted to the risk index synthesis unit 330.

[0109] The risk index synthesis unit 330 receives the autocorrelation coefficient sequence and variance sequence, and analyzes their changing trends to determine early warning signals. When system stability decreases, i.e., when it approaches the critical transition point, its external manifestation is that both the autocorrelation coefficient and variance show a monotonically increasing trend over time.

[0110] To quantify this trend, the risk index synthesis unit 330 is used at each time point t. j This unit calculates the strength of the trend of the AC1(t) and Var(t) sequences over a period of historical data prior to the current time point. In one specific implementation, this unit calculates the Kendall's taurank correlation coefficient to obtain the trend coefficients AC1 and τ of the AC1 sequence. Var (t j ).

[0111] This unit synthesizes a critical risk index I based on these two trend coefficients. risk (t j In one implementation, the index is calculated by the following formula:

[0112]

[0113] If and only if and τ Var (t j When both are positive, I risk (t j A positive value indicates a risk that the system is losing stability. The magnitude of this index reflects the strength of the risk signal.

[0114] Finally, the risk index synthesis unit 330 outputs a critical risk index I in time series form. risk (t), and transmit it to the collaborative diagnosis and report generation module 400 for multimodal collaborative analysis.

[0115] Please refer to the appendix. Figure 5 , Figure 5 This is a structural block diagram of a collaborative diagnosis and report generation module 400 according to an embodiment of the present invention. The module may include a collaborative diagnosis unit 510 and a report generation unit 520.

[0116] The collaborative diagnosis and report generation module 400 receives the analysis results from the aforementioned modules. Specifically, its collaborative diagnosis unit 510 receives and integrates three types of heterogeneous information: the critical risk index I generated by the causal inference and risk assessment module 300. risk (t) and dynamic causal networks, as well as topological fingerprints generated by the dynamics and topology analysis module 200.

[0117] In one specific implementation, the collaborative diagnostic unit 510 executes a multi-level collaborative diagnostic process to improve the accuracy of early warning determination.

[0118] First, the collaborative diagnostic unit 510 is based on the critical risk index I. risk (t) Perform Level 1 risk assessment. When the risk index exceeds a preset risk threshold Th at consecutive time points. risk At this time, the system generates an initial warning signal, indicating that the system stability may be declining.

[0119] Next, the collaborative diagnostic unit 510 cross-validates the initial warning signal using topological fingerprinting. This unit analyzes the lifetime of the topological features represented in the persistence graph. If, during the same period of rising risk index, long-lifetime 1D topological features (i.e., loop structures) appear or significantly increase in the persistence graph, it indicates a qualitative change in the system's state-space trajectory. This finding is used to confirm the validity of the initial warning signal.

[0120] After the warning signal is confirmed, the collaborative diagnostic unit 510 uses a dynamic causal network to trace the mechanism. The collaborative diagnostic unit 510 analyzes regions in the network where node centrality or edge weights change significantly during the risk accumulation period. Through this analysis, the collaborative diagnostic unit 510 identifies key driving molecules that dominate the changes in system state during this period, or identifies abnormally activated regulatory loops.

[0121] The collaborative diagnostic unit 510 integrates the results of the above three-level analysis to generate a comprehensive diagnostic conclusion that includes the final risk level, topological state description and a list of key causal paths, and then transmits it to the report generation unit 520.

[0122] After receiving the comprehensive diagnostic conclusion output by the collaborative diagnostic unit 510, the collaborative diagnostic and report generation module 400 integrates and formats the conclusion and related supporting data through its report generation unit 520 to generate a comprehensive early warning report.

[0123] In one specific implementation, the comprehensive early warning report generated by the report generation unit 520 includes the following core components:

[0124] The first part is the overall risk assessment. Based on the final judgment of the collaborative diagnostic unit 510, this part provides a clear risk level classification, such as "low risk", "medium risk" or "high risk warning".

[0125] The second part is dynamic trajectory analysis. This part presents the topological fingerprints generated for each individual in a visual manner, which in one implementation is a persistent graph. The report annotates specific topological structures in the graph that reflect the characteristics of the system's state evolution, for example, identifying and explaining long-lifetime topological features that indicate the occurrence of closed loops or increased structural complexity in the system trajectory.

[0126] The third part traces the causal mechanism. This section presents the key sub-network structure diagrams extracted from the dynamic causal network. The diagrams represent core molecular features as nodes and their causal interactions as directed edges. In one implementation, the diagrams visually distinguish molecules and paths where edge weights or node centralities change significantly during the risk accumulation phase.

[0127] In addition, the report includes a data appendix containing basic data charts to support the above conclusions, such as the critical risk index I. risk The graphs of (t) versus time, the first-order autocorrelation coefficient AC1(t), and the variance Var(t) versus time are provided for reference.

[0128] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A tumor early screening and subtyping warning system based on multi-omics data association analysis, characterized in that, include: The data acquisition and preprocessing module is used to acquire and process longitudinal multi-omics data from an individual; The dynamics and topology analysis module is used to generate an evolutionary trajectory point cloud for the individual based on the longitudinal multi-omics data, and to generate a topological fingerprint based on the evolutionary trajectory point cloud; The causal inference and risk assessment module is used to infer a dynamic causal network based on the longitudinal multi-omics data and calculate a critical risk index. The collaborative diagnosis and report generation module is used to integrate the topological fingerprint, the dynamic causal network, and the critical risk index to generate a comprehensive early warning report.

2. The tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 1, characterized in that, The dynamics and topology analysis module includes: The state space reconstruction unit is used to embed the longitudinal multi-omics data through a time delay to reconstruct and generate the evolutionary trajectory point cloud. The reconstruction parameter optimization unit is used to determine the embedding dimension and time delay parameters for the state space reconstruction unit before performing time delay embedding. The topological fingerprint generation unit is used to analyze the evolutionary trajectory point cloud to generate the topological fingerprint.

3. The tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 1, characterized in that, The causal inference and risk assessment module includes: A dynamic causal network construction unit is used to analyze the time series relationships between different molecular features in the longitudinal multi-omics data in order to infer the dynamic causal network; A critical slowing computation unit is used to calculate one or more statistical indicators of the longitudinal multi-omics data within a sliding time window; The risk index synthesis unit is used to synthesize the critical risk index based on the changing trends of the one or more statistical indicators.

4. The tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 1, characterized in that, The collaborative diagnosis and report generation module includes: A multimodal information fusion unit is used to perform collaborative diagnosis of the critical risk index, the topological fingerprint, and the dynamic causal network; The risk tracing and classification unit is used to trace the source and potential classification of risks based on the driving nodes and paths in the dynamic causal network. The visualization report generation unit is used to generate the comprehensive early warning report, which includes risk curves, topology diagrams, and network diagrams.

5. The tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 1, characterized in that, The data acquisition and preprocessing module includes: The longitudinal data interface unit is used to receive and manage the longitudinal multi-omics data from different time points; The data standardization unit is used to normalize or standardize the longitudinal multi-omics data. The data quality control unit is used to perform missing value processing and batch effect correction on the longitudinal multi-omics data.

6. The tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 2, characterized in that, The topological fingerprint generation unit is used to analyze the evolutionary trajectory point cloud using a persistent cohomology algorithm, and quantize the lifecycle of the topological features of the evolutionary trajectory point cloud into a persistent graph or persistent barcode as the topological fingerprint.

7. The tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 2, characterized in that, The reconstruction parameter optimization unit determines the time delay parameter using the average mutual information method and the embedding dimension parameter using the pseudo-nearest neighbor algorithm.

8. The tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 3, characterized in that, The critical slowing calculation unit calculates the first-order autocorrelation coefficient and variance of the longitudinal multi-omics data within the sliding time window as one or more statistical indicators, and determines that the synchronous and continuous upward trend of the first-order autocorrelation coefficient and variance is a warning signal of system instability.

9. The tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 3, characterized in that, The dynamic causal network construction unit uses a convergent cross-mapping algorithm to infer the dynamic causal network.

10. A tumor early screening and subtyping warning system based on multi-omics data association analysis according to claim 4, characterized in that, The multimodal information fusion unit is used for: The critical risk index is used as the primary criterion for determining the risk level. The changes in topological features in the topological fingerprint are used as a secondary criterion for confirming system stability; The dynamic causal network is used as a three-level judgment basis for risk tracing and classification.

Images