Machine learning-based method for quantifying endogenous circadian rhythm disorder of atherosclerosis

Abstract

The invention discloses an atherosclerosis endogenous circadian rhythm disorder quantification method based on machine learning, and relates to the technical field of machine learning diagnos.The method comprises the steps that S1, bulk RNA-seq and scRNA-seq normalization values are calculated to obtain the average expression quantity Ei, S2, a random sampling strategy is adopted, a gene i is divided into 100 expression partitions according to the average expression quantity Ei of the gene i, and the expression partitions are divided into 100 expression partitions according to the expression quantity Ei of the gene i; the method comprises the following steps: S1, identifying CTS frequency in each partition, S3, defining a given normalized expression matrix M as data without excessive migration signals, S4, creating a random score SRandom, and S5, quantizing CockProAS. According to the method, the endogenous circadian rhythm disorder characteristics of patients with atherosclerosis diseases can be quantified, comprehensive evaluation is carried out according to clinical and biological backgrounds, and dynamic development of atherosclerosis under the influence of circadian rhythm system disorder is reflected.

Description

Technical Field

[0001] The present invention relates to the field of machine learning diagnostic technology, and more specifically to a method for quantifying endogenous circadian rhythm disturbance in atherosclerosis based on machine learning. Background Art

[0002] Throughout history, humans have undergone major changes in circadian behavioral patterns, with consequences for populations such as night shift workers and those with sleep disorders. Clinical studies have highlighted the role of diurnal variations in cardiovascular parameters in the pathophysiology and pathogenesis of cardiovascular complications. For example, in humans, there is a remarkable consistency in the frequency of thromboembolic and cardiovascular events, peaking in the early morning. Emerging evidence suggests that circadian variations critically affect hematopoiesis, inflammatory status, and lipid supply, all of which are systemic drivers of atherosclerosis, suggesting that the intrinsic biological clock is a key regulator of the atherosclerotic process.

[0003] Asynchrony of the circadian clock and the associated inflammatory processes, endothelial dysfunction, and lipid imbalance may drive the progression of atherosclerotic lesions. In these lesions, circadian rhythms regulate cellular dysfunction and behavior, but the detailed mechanisms linking circadian clock disruption to plaque vulnerability are not fully understood. Understanding these mechanisms is critical for developing targeted preventive strategies and therapies to address the effects of circadian disruption in atherosclerotic disease.

[0004] In animal models, molecular clock perturbations have been shown to contribute to cardiovascular disease, but no animal model fully replicates or represents human pathophysiology. Omics approaches hold promise for identifying biomarkers and providing mechanistic insights into the association between circadian rhythm perturbations and risk of atherosclerosis. Current understanding relies on epidemiological studies and animal models, such as complete knockout mouse models of molecular clock transcription factor perturbations. Furthermore, the lack of tools that can capture and accurately characterize transient states of circadian rhythm perturbations necessitates detailed analysis of the links between molecular features of circadian rhythm perturbations and their impact on human physiology.

[0005] Therefore, it is necessary to propose a machine learning-based method for quantifying endogenous circadian rhythm disturbance in atherosclerosis to solve the above problems. Summary of the invention

[0006] The purpose of the present invention is to solve the problem of lack of tools that can capture and accurately describe the transient state of circadian rhythm disorders.

[0007] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions: A method for quantifying endogenous circadian rhythm disturbance in atherosclerosis based on machine learning, comprising the following steps: Step S1: Calculation Normalized value to get the average expression : , Where: is the given normalized expression matrix, is the number of samples, For genes in cells, , the above formula is , and the following formula is ; Step S2: random sampling strategy is adopted to select The average expression of Gene It is divided into 100 expression partitions, and the frequency of identified CTS in each partition is counted, which is called , according to random sampling The number of times, randomly select the feature gene K from each partition, ensuring that the total number of random feature genes K matches the number of CTS, and this random process is repeated 1000 times; Step S3: normalize the given expression matrix Defined as data without excessive migration signals: , In the formula, It's genetic In cells The central expression in ; Step S4: Create random scores , as the random feature gene mentioned above × Average of 1000 random feature samples, CTS score According to each cell The average value of K CTS of random feature genes determined by the central expression data; Step S5: Quantification The characteristic of is a normal distribution or a mixture of normal distributions, calculated as follows: .

[0008] Furthermore, the detection of the CTS comprises the following steps: Step S11, using the cell data obtained from atherosclerotic plaques of symptomatic and asymptomatic patients, using the k-nearest neighbor algorithm to identify similar cell groups that need to be aggregated, calculating the average expression levels of these cells, and obtaining a meta-cell gene expression matrix; Step S12, constructing a gene-gene correlation adjacency matrix based on the meta-cell gene expression matrix, and selecting the lowest soft power threshold with a scale-free topological model fit of ≥ 0.8; Step S13, converting the adjacency matrix into a co-expression network to obtain a topological overlap matrix and 1-TOM, defining a module that is highly correlated with the progression of atherosclerotic plaques from early to late stages as a circadian transcriptome driver CTD that accelerates plaque progression; Step S14, atherosclerosis usually goes through four different disease state stages: relatively normal state, pre-disease state, progress state and decline state, and static snapshots of different stages of the patient's atherosclerosis progression are integrated into a disease state matrix; Step S15, using CTD as input, removing low-expression transcription noise, and constructing a network graph for each disease state matrix, where the nodes are CTDs and the edges are connected genes with significantly correlated expression levels; Step S16, using the random walk algorithm, find strongly connected components in each network, divide the co-expressed CTDs into gene modules of each disease state matrix as input for subsequent CTS search targets; Step S17: using a dynamic network biomarker model to quantify each gene module in each disease state, thereby detecting candidate CTSs from different gene modules in each disease state.

[0009] Compared with the prior art, the present invention has the following beneficial effects: This invention can quantify the endogenous circadian rhythm disorder characteristics (ClockProAS) of patients with atherosclerotic diseases and conduct a comprehensive assessment based on clinical and biological backgrounds, reflecting the dynamic development of atherosclerosis under the influence of circadian rhythm system disorders. It describes the association between personalized circadian rhythm molecular characteristics and subsequent pathophysiological mechanisms, disease progression and remission, and other clinical manifestations. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 :High-dimensional network analysis of the circadian system in the progression of atherosclerosis; (A) Single-cell projections of atherosclerotic plaques in symptomatic and asymptomatic patients.

[0011] (B) Depiction of expression matrix density (1, sparsity) for single-cell (sc), pseudobulk (pb), and meta-cell matrices.

[0012] (C) Two-dimensional UMAP embedding of the topological overlap matrix (TOM) showing the circadian co-expression network from plaques.

[0013] (D) Systems-level association between circadian module genes and plaque exacerbation.

[0014] (E) Overlap analysis comparing circadian rhythm-related gene sets and identified modules using Fisher’s exact test.

[0015] (F) Differential module genes (DME) detection between asymptomatic and symptomatic plaques.

[0016] (G) Identification, summary, and visualization of high-order functional themes of four circadian modules (M1, M2, M7, M8).

[0017] Figure 2 : Circadian rhythm dysregulation and dynamic changes in the progression of atherosclerosis; (A) Overview of the dynamic risk assessment procedure combining DNB and Ic theories.

[0018] (B) Detecting CTSs from multiple gene modules in each state by combining the DNB model with optimization and network partitioning methods.

[0019] (C) The Ic model validates the robustness of the criticality patterns and CTS captured by DNB.

[0020] (D) Global differences in CTS gene expression between late and early plaque tissues were explored in the GSE28829 cohort.

[0021] (E) Overall differences in CTS gene expression between ruptured plaque tissue and stable plaque tissue in the GSE41571 cohort.

[0022] Figure 3 :Robustness of ClockProAS in quantifying circadian misalignment; (A) Circadian rhythm detection of ClockProAS in different samples.

[0023] (B) Differential circadian rhythm analysis revealed.

[0024] (C) Distribution of ClockProAS values ​​in four independent datasets.

[0025] (D) Mendelian randomization analysis showed.

[0026] (E) Distribution of ClockProAS values ​​in three different shift work paradigms.

[0027] (F) ClockProAS values ​​of circadian rhythm mutant mice and wild-type mice were analyzed and further divided into atherosclerosis subgroup and control subgroup.

[0028] (G) The ClockProAS values ​​of circadian rhythm mutant mice and wild-type mice were analyzed and divided into myocardial ischemia subgroup, ischemia-reperfusion subgroup, preconditioning subgroup and control subgroup.

[0029] Figure 4: Macro-clinical interpretability behind circadian rhythm dysregulation; (A) Line graph showing the log hazard ratios (HRs) for the association between ClockProAS and major adverse cardiovascular events (MACE), as well as the HRs (blue line) and 95% confidence intervals (shaded area).

[0030] (B) Kaplan-Meier curves of STEMI-related MACE in GSE21545, with P values ​​from the log-rank test.

[0031] (C) Heat map of the relationship between circadian gene profiles and clinical variables in cohort GSE90074.

[0032] (DI) Line graphs and box plots depict the mean and distribution of ClockProAS values ​​in the six cohorts, highlighting the association with atherosclerosis severity, progression, and remission timeline.

[0033] (J) 3D plot depicting the interaction of ClockProAS, age, and Duke Coronary Artery Disease Index, and Gantt chart of stenosis degree, circulating inflammation level, and Duke Coronary Artery Disease Index in deciles of ClockProAS.

[0034] (K) Evaluation of the robustness and generalizability of ClockProAS. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] See also Figure 1-4 , a machine learning-based method for quantifying endogenous circadian rhythm disturbance in atherosclerosis, consists of four steps; Deciphering transcriptomic drivers of atherosclerosis progression in the circadian system through high-dimensional network analysis Using data obtained from atherosclerotic plaques from symptomatic and asymptomatic patients, we reconstructed the circadian spectrum of atherosclerosis at the single-cell level.

[0037] The k-nearest neighbor method was used to calculate the average expression of 50 neighboring cells, and highly similar cells were collapsed into "meta-cells" to reduce sparsity and create a modular design that can be used for network analysis under different patch conditions ( Figure 1A, EC, endothelial cell; FC, fibrochondrocyte; SMC, vascular smooth muscle cell; SEM, SMC-derived intermediate cell; Macro, macrophage; Mono, monocyte; DC, dendritic cell). This strategy significantly reduced the sparsity of the circadian meta-cell matrix for each cell type compared to the single-cell matrix ( Figure 1 B , Depiction of expression matrix density (1, sparsity) for single-cell (sc), pseudobulk (pb), and meta-cell matrices, considering different K values ​​in each cell type).

[0038] Subsequently, hierarchical clustering revealed nine circadian co-expression modules ( Figure 1 C, Nodes represent individual genes, colored by module assignment, and edges represent co-expression links between genes and module hub genes. Point sizes are scaled by gene-based connectivity (kME), and network edges are downsampled for visual clarity). All of these modules are significantly involved in the progression of human plaques from an asymptomatic to a symptomatic state ( Figure 1 D, Correlations were calculated for each cell type and integrated to elucidate the overall context of the circadian system in atherosclerosis). Figure 1 E, Similarity was assessed by the Jaccard similarity coefficient (Jaccard index, X indicates non-significant overlap (FDR>0.05)). Among these modules, four were significantly enriched in the circadian rhythm-related gene set. Differential module gene (DME) analysis further highlighted their unique co-expression patterns during the transition from asymptomatic to symptomatic state ( Figure 1 F, to discover changes in the genes in the modules). Therefore, these four modules (577 genes) are expected to reflect circadian transcriptomic drivers (CTDs) that accelerate plaque progression.

[0039] To understand the higher-order molecular phenotypes behind CTD, we introduced the vissE system framework and identified eight priority biological functional themes in the progression of atherosclerosis. These themes were mainly related to circadian signaling disruption pathways, extracellular matrix (ECM) organization, and antigen presentation ( Figure 1G, Left: Overlapping network made using genome similarity data of four modules and MsigDB gene features, annotated using fold change (logFC). The Qusage algorithm identified significantly dysregulated gene sets between symptomatic and asymptomatic patches, and similarity was calculated using the Jaccard index. A graphical clustering method (walktrap algorithm) identified eight higher-order biological functional communities. Right: VissE autonomously assessed the biological similarity of each community using a text mining approach. Frequency analysis of short descriptions of genomes (adjusted using inverse document frequency) was used to assess recurring biological themes in the communities and presented as word clouds).

[0040] Corresponding technical methods: High Dimensional Network Analysis (1) Cell construction: Metacells are small groups of similar cells derived from the same biological sample. The k-nearest neighbor algorithm is used to identify similar cell groups that need to be aggregated. Then the average expression or sum of expression of these cells is calculated to obtain the meta-cell gene expression matrix. Compared with the original expression matrix, the sparsity of the meta-cell expression matrix is ​​greatly reduced while retaining the heterogeneity of cells, so it is more suitable for use. The reason for using meta-cells instead of original single cells was originally because the correlation network method is sensitive to data sparsity.

[0041] (2) Co-expression network analysis: The meta-cell expression matrix was used for network analysis. A gene-gene correlation adjacency matrix was constructed based on the meta-cell expression matrix, and the lowest soft power threshold with a scale-free topological model fit of ≥ 0.8 was selected. This careful thresholding effectively reduced the noise in the correlation matrix, emphasizing robust connections while weakening weak connections.

[0042] The adjacency matrix was converted into a co-expression network to obtain the topological overlap matrix (TOM) and 1-TOM, which reflect the similarity and dissimilarity between genes, respectively. The modules that were highly associated with the progression of atherosclerotic plaques from early to late stages were identified as circadian transcriptome drivers (CTDs) that accelerate plaque progression.

[0043] 2. Exploring the impact of the circadian system on the evolution of atherosclerosis Atherosclerosis typically progresses through four different disease states: relatively normal (pre-exacerbation), pre-disease (critical), progressive (exacerbation), and decline. Therefore, static snapshots of different stages of a patient's atherosclerosis progression were composited into the video, providing a personalized dynamic readout of an individual's disease risk state through the lens of circadian rhythms.

[0044] By fuzzy mean clustering, the perturbation pattern of CTD in the progression trajectory of different stages of acute coronary syndrome (ACS) was revealed: stable control group; unstable angina (UA); non-ST-segment elevation myocardial infarction (NSTEMI); STEMI; after percutaneous coronary intervention (PCI). It is worth noting that the expression dynamics of CTD does not follow a gradual monotonic path, but shows a nonlinear sharp transition.

[0045] The key to explaining the biological impact of circadian rhythms on the pathogenesis of atherosclerosis is to identify critical periods, i.e., pre-disease states. Two commonly used models based on tipping point theory were enhanced using optimization and network decomposition techniques to design a dynamic risk assessment procedure to explore malignant phase transitions and critical transition signals (CTS) ( Figure 2 A. Despite specific differences in biological processes or symptoms across diseases, disease progression can generally be divided into different states: relatively normal (pre-exacerbation), pre-disease (borderline), progression (exacerbation), and decline. While there are clear differences between normal and progressive states, the demarcation between relatively normal and pre-disease states is subtle because the latter is inherently part of the former. Therefore, traditional molecular signatures may not adequately describe systemic changes during disease evolution due to their static nature. Using a cohort, the goal was to explore the association between circadian oscillations and transitions between atherosclerotic states). The program incorporated CTD expression profiles across five cross-sections of ACS development.

[0046] For each disease state matrix, a network graph was constructed in which nodes were CTDs and edges were connected genes with significantly correlated expression levels.

[0047] To identify strongly connected components in specific graphs within each network map, the co-expressed CTDs were divided into modules for each disease state matrix, which served as input for subsequent CTS search targets.

[0048] When each gene module in each cross section was quantified according to the dynamic network biomarker (DNB) model, a strong tipping point signal before progression was captured in the NSTEMI stage ( Figure 2 B). The top two significant modules of the critical point were identified as CTS candidate genes, including 251 genes ( Figure 2B, Bar graph showing the DNB scores (combined index (CI)) for five ACS snapshots. The horizontal dashed line indicates the cutoff of 0.5, resulting in two modules as CTSs, with the module size noted at the top. Comparison of observed (red dots) and simulated (boxes, based on CTS genes) CTS DNB scores revealed that the DNB model was significant (P < 0.001). The critical index (IC) model further validated the dramatic changes in the co-expression relationships within CTSs and between CTSs and other molecules during the development of atherosclerosis ( Figure 2 C, Random Ic scores were calculated by shuffling CTS genes. The figure shows the comparison of the identified Ic scores (red and blue) as the critical points at different stages with their corresponding size-controlled random scores (grey).

[0049] Subsequently, the Euclidean distance was used to measure the overall changes in CTS expression between and within late-stage plaque tissue and early-stage plaque tissue. The expression distance between late-stage plaque tissue and early-stage plaque tissue was the largest, indicating that there are intrinsic circadian rhythm differences within atherosclerotic tissue ( Figure 2 D, Euclidean expression distances were calculated between late and early plaques (blue), within late tissue samples (sky blue), and within early tissue samples (gray). Insets summarize the average distances between tissue pairs as a percentage of the average distances between late and early plaques. Similar results were observed for ruptured and stable plaques ( Figure 2 E). In conclusion, circadian-based CTS can reflect the dynamic progression status of atherosclerosis and provide potential insights into the regulatory oscillations of circadian biological systems during this critical period.

[0050] Technical methods used: Transcriptome tipping point analysis can be widely applied to gain insights into the plasticity, heterogeneity, and phenotypic changes of dynamic biological systems.

[0051] (1) Feature pre-selection: Using CTD as input, we remove low-expression transcription noise and reduce downstream computational burden.

[0052] (2) Network partition: Correlations between genes reveal gene regulatory networks. When perturbations exceed the equilibrium threshold, critical transition events occur. It is crucial to determine which genes are most influential in these co-expression networks. A network graph was constructed for each disease state matrix with CTDs as nodes. Edges connected genes with correlated expression (Pearson correlation FDR < 0.05). A random walk algorithm was used to find strongly connected components (or gene modules) in the network graph. The co-expressed CTDs were then divided into modules by state, which became the basis for identifying critical transition signal (CTS) candidates in the subsequent stage.

[0053] (3) Identify candidate CTS: Using the Dynamic Network Biomarker (DNB) model, each gene module in each disease state or snapshot is quantified, so that candidate CTSs can be detected from different modules in each state. This step will output multiple candidate CTSs (modules with significant significance in the DNB scoring system) instead of just the one with the highest score.

[0054] (4) Importance Technical Assessment: To ensure robustness and reliability, the last step considers the common CTS states recaptured by the DNB model and the Ic model. To improve the accuracy of Ic, an advanced estimation method, “correlation shrinkage”, is introduced. The newly defined score is named Ic.shrink. Important CTSs meet the following criteria: a. The DNB score was significantly increased, indicating an increase in the interconnectedness between genes within the identified CTS-related groups.

[0055] b. The Ic.shrink score of this CTS is significantly higher, indicating that the critical transition occurs in the same gene cluster.

[0056] 3. Based on the above analysis results, develop, build and verify the degree of quantification of endogenous circadian clock disturbance (ClockProAS) To assess the extent of endogenous circadian clock perturbations in single cells or bulk samples, a ClockProAS quantification system was designed that can adjust the signal-to-noise ratio of different genes and cells based on their expression profiles.

[0057] calculate Normalized value to get the average expression : , Where: is the given normalized expression matrix, is the number of samples, For genes in cells, , the above formula is , and the following formula is ; A random sampling strategy was used to select The average expression of Gene It is divided into 100 expression partitions, and the frequency of identified CTS in each partition is counted, which is called , according to random sampling The number of times, randomly select the feature gene K from each partition, ensuring that the total number of random feature genes K matches the number of CTS, and this random process is repeated 1000 times; The given normalized expression matrix Defined as data without excessive migration signals: , In the formula, It's genetic In cells The central expression in ; Creating a random score , as the random feature gene mentioned above × Average of 1000 random feature samples, CTS score According to each cell The average value of K CTS of random feature genes determined by the central expression data; Quantification The characteristic of is a normal distribution or a mixture of normal distributions, calculated as follows: .

[0058] To validate the performance and robustness of ClockProAS, 16 independent datasets involving different circadian patterns were used. Significant circadian rhythmicity was detected under various conditions using a likelihood-based approach. This rhythmicity was observed in the blood of healthy humans, the suprahypothalamic nucleus of healthy primates Papio-Anubis, and the heart and liver tissues of C57 / BL6 and C57Bl / 6J mice ( Figure 3 A, ClockProAS demonstrated precise detection of circadian rhythmicity within circadian time in different samples, including blood of healthy humans, the suprathalamic nucleus of Papio Anubis, and cardiac tissue of C57 / BL6 mice.

[0059] In addition, ClockProAS can successfully distinguish different rhythmic patterns between circadian rhythm disorders and relatively normal states ( Figure 3 B, Basal levels of ClockProAS are higher when circadian rhythms are disrupted relative to a more stable state (induced by sleep conditions, transgenics, or reversed feeding schedules). Increased basal levels of ClockProAS were observed in sleep-restricted versus well-rested humans, in circadian clock mutant mice versus wild-type mice, and in mice with nighttime versus daytime food restriction ( Figure 3B, shows association with circadian disruption). In nine other conditions, samples subjected to circadian disruption treatment (CDT) showed higher ClockProAS than samples under relatively normal conditions ( Figure 3 C).

[0060] Shift work may lead to circadian rhythm disorders and is therefore associated with an increased risk of cardiovascular disease. Through Mendelian randomization analysis, it was found that shift work had a significant causal effect on myocardial infarction (MI) (Beta: 1.769, 95% CI: [1.091, 2.868], P value: 0.02) ( Figure 3 D, Shift work has a significant causal effect on MI, with Beta and Odds ratios in the left and right columns, respectively. IVW, inverse variance weighting). The ClockProAS levels of different shift work patterns differed significantly, with night shift workers having the highest ClockProAS levels, followed by day shift workers, and then non-shift work control group workers ( Figure 3 E).

[0061] Furthermore, ClockProAS levels were higher in circadian rhythm mutant mice compared with wild-type mice, and in atherosclerotic samples compared with healthy samples ( Figure 3 F, ClockProAS values ​​of circadian rhythm mutant mice and wild-type mice were analyzed and further divided into atherosclerosis subgroup and control subgroup. A similar trend was observed in another dataset, where mice with myocardial ischemia showed the highest ClockProAS levels ( Figure 3 G). These results suggest that ClockProAS can effectively determine the status of circadian rhythm disturbance.

[0062] 4. Quantification system (ClockProAS) provides macro clinical insights into circadian rhythm disorders The initial evaluation focused on the predictive accuracy of ClockProAS for clinical outcomes. Restricted cubic spline models confirmed a significant linear prognostic relationship between ClockProAS and major adverse cardiovascular events (MACE) ( Figure 4 A), Univariate Cox regression analysis further confirmed that ClockProAS was a significant predictor of MACE (HR=1.883, P<0.05, Figure 4 A, The analysis used a Cox-type additive hazard regression model with restricted cubic spline regression, with 0.0144 ClockProAS as the reference. Kaplan-Meier curves showed that the incidence of MACE in patients with chronic coronary syndrome (CCS) with elevated ClockProAS was increased (log-rank p < 0.01, Figure 4 B).

[0063] In addition, the correlation between ClockProAS and different types of clinical variables, including clinical manifestations, comorbidities, inflammatory factor levels, and severity of atherosclerosis, was determined at the molecular level. Hierarchical clustering of Spearman correlations showed that the ClockProAS gene profile was closely associated with the severity of coronary artery stenosis and obstructive CAD ( Figure 4 C, Genes grouped by signature (ClockProAS, CTDs). The color range of the heatmap entries shows the p-value for each association (enrichment and depletion) on a negative logarithmic scale. ClockProAS levels increased with increasing coronary artery stenosis (grade 1: less than 10% stenosis in all major coronary arteries; grade 2: 10% to 70% stenosis in one vessel, excluding the left main coronary artery; grade 3: more than 70% stenosis in one vessel, excluding the left main coronary artery; grade 4: more than 70% stenosis in one vessel, excluding the left main coronary artery): grade 3: >70% stenosis in one vessel; grade 4: >70% stenosis in two vessels; grade 5: >70% stenosis in three vessels; grade 6: >70% stenosis in three vessels: >P < 0.001, Figure 4 D). This trend was also supported by another classification standard in another dataset (p<0.0001, Figure 4 E).

[0064] In addition, ClockProAS levels fluctuate with the onset and remission of various diseases. Increased ClockProAS often indicates that the disease is about to become active. ClockProAS levels are significantly increased in the acute phase of the disease, especially in STEMI patients, followed by NSTEMI, AMI after PCI, UA and healthy controls (p<0.001, Figure 4 F). It is noteworthy that when admitted to hospital for myocardial infarction, patients' ClockProAS levels were significantly higher than those of stable CCS patients, but after mechanical reperfusion or successful stent implantation, patients' ClockProAS levels decreased and continued to decrease throughout the recovery period ( Figure 4 G-4I). The higher the ClockProAS level, the greater the likelihood of coronary artery restenosis after PCI ( Figure 4K, Top: ClockProAS was calculated for more than 1,400 patients with atherosclerosis in 10 cohorts. The gray dashed line represents the median APVSLevel of all samples. Left: The bar chart shows the proportion of ClockProAS deciles in the 10 cohorts. The right graph shows the proportion of samples in the ClockProAS-enhanced and ClockProAS-reduced categories in the 10 cohorts, with the ClockProAS median as the dividing line. Further analysis showed that as the ClockProAS deciles increased, the degree of stenosis (p<0.0001), Duke Coronary Artery Disease Index (p<0.0001), and circulating inflammation levels (p<0.001) also gradually increased ( Figure 4 J).

[0065] Analysis of more than 1,400 patients with atherosclerosis in 10 bulk cohorts showed that the distribution pattern of ClockProAS was consistent and the proportions of each group were comparable, indicating that the quantification system has stable universality ( Figure 4 K).

[0066] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. The patent protection scope of the present invention shall be based on the claims. All equivalent structural changes made using the contents of the description and drawings of the present invention should also be included in the protection scope of the present invention.

Claims

1. A method for quantifying endogenous circadian rhythm disturbance in atherosclerosis based on machine learning, characterized in that: The following steps are involved: Step S1: Calculate the normalized values ​​of bulk RNA-seq and scRNA-seq to obtain the average expression level E i Where: M is the given normalized expression matrix, N is the number of samples, i is the gene in the cell, q = 1,...,N, the above formula is bulkRNA-seq, and the following formula is scRNA-seq; Step S2: Using a random sampling strategy, according to the average expression level E of gene i i Divide gene i into 100 expression partitions and count the frequency of identified CTS in each partition, called B CTS , according to random sampling B CTS The number of times, randomly selects the feature gene K from each partition, ensuring that the total number of random feature genes K matches the number of CTS, and this random process is repeated 1000 times; Step S3, define the given normalized expression matrix M as data without excessive migration signals: In the formula, Z i,q is the central expression of gene i in cell q, q = 1,...,N; Step S4: Create a random score S random , as the average value of the above random feature gene K×1000 random feature samples, CTS score S CTS It is the average value of K CTS of random feature genes determined based on the central expression data of each cell q; Step S5, quantify ClockProAS, the characteristic of ClockProAS is a normal distribution or a mixed distribution of normal distributions, and the calculation method is as follows: ClockProAS=S random -S CTS 。 2. The method for quantifying endogenous circadian rhythm disturbance in atherosclerosis based on machine learning according to claim 1, characterized in that: The detection of the CTS comprises the following steps: Step S11, using the cell data obtained from atherosclerotic plaques of symptomatic and asymptomatic patients, using the k-nearest neighbor algorithm to identify similar cell groups that need to be aggregated, calculating the average expression levels of these cells, and obtaining a meta-cell gene expression matrix; Step S12, constructing a gene-gene correlation adjacency matrix based on the meta-cell gene expression matrix, and selecting the lowest soft power threshold with a scale-free topological model fit of ≥ 0.8; Step S13, converting the adjacency matrix into a co-expression network to obtain a topological overlap matrix and 1-TOM, defining a module that is highly correlated with the progression of atherosclerotic plaques from early to late stages as a circadian transcriptome driver CTD that accelerates plaque progression; Step S14, atherosclerosis usually goes through four different disease state stages: relatively normal state, pre-disease state, progress state and decline state, and static snapshots of different stages of the patient's atherosclerosis progression are integrated into a disease state matrix; Step S15, using CTD as input, removing low-expression transcription noise, and constructing a network graph for each disease state matrix, where the nodes are CTDs and the edges are connected genes with significantly correlated expression levels; Step S16, using the random walk algorithm, find strongly connected components in each network, divide the co-expressed CTDs into gene modules of each disease state matrix as input for subsequent CTS search targets; Step S17: using a dynamic network biomarker model to quantify each gene module in each disease state, thereby detecting candidate CTSs from different gene modules in each disease state.

Images