A network analysis method for mass spectrometry streaming data mining
The SCANCell analysis method converts mass spectrometry flow data into DA networks between cell clusters, solving the problem that the existing technology cannot mine interaction patterns, providing new understanding of disease pathogenesis, and realizing the identification and potential of disease-specific interaction patterns Identification of drug targets.
Patent Information
- Application Number
- CN202111185696.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-12
- Publication Date
- 2024-12-20
- Estimated Expiration
- 2041-10-12
AI Technical Summary
Existing high-dimensional data analysis methods cannot effectively mine the interaction patterns between hidden cell clusters in mass spectrometry flow data, and it is difficult to understand the pathogenesis of multifactorial complex diseases such as systemic lupus erythematosus.
A SCANCell analysis method was developed to convert high-dimensional mass spectrometry flow data into DA networks between cell clusters, and to construct topological structures through cell protein expression and cluster abundance, and to quantify interactions between EMD and partial mutual information to construct And visualize the DA network to identify disease-specific abnormal interaction patterns.
Successfully transforming hidden information into interaction patterns between cell clusters provides a new dimension to understand disease pathogenesis, can dig out abnormal interaction patterns between healthy and disease samples from the network, revealing the onset of systemic lupus erythematosus mechanisms and provide new angles for the identification of potential drug targets.
Smart Images
Figure CN113851193B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data analysis models, and in particular to a network analysis method for mass spectrometry streaming data mining. Background Art
[0002] Systemic lupus erythematosus (SLE) is a systemic autoimmune disease that can cause damage to multiple organs, including the skin, joints, kidneys, and central nervous system. The onset of the disease is very variable, including dermatology, nephrology, neurology, hematology, or rheumatology, which makes accurate diagnosis and clinical treatment challenging. Despite these challenges, advances in the understanding of the pathogenesis of SLE have facilitated the treatment of clinical patients.
[0003] Advances in the study of the pathogenesis of systemic lupus erythematosus have emphasized the contribution of type I interferon (IFN) and multiple immune cell types to autoantibody production and tissue damage. Chronic IFN production and its effect on immune system activation have become core mechanisms of SLE. IFN is produced after activation of innate plasmacytoid dendritic cells (pDCs) mediated by Toll-like receptors (TLRs). Excessive activation of B cell and T cell immune responses and loss of immune tolerance to self-antigens are also one of the causes of SLE. T follicular helper cells can promote the differentiation of B cells that produce autoantibodies. Regulatory T (Treg) cells can suppress immune responses and maintain the immune system, but the abundance of cells is reduced in SLE. Although cumulative studies have laid the foundation for a preliminary understanding of the disease, due to the cellular complexity and functional diversity of the immune system, it is necessary to use high-dimensional single-cell technology to study the pathogenesis of systemic SLE.
[0004] Mass spectrometry flow cytometry (Cytometry by time-of-flight, CyTOF) has greatly increased the number of cellular parameters that can be simultaneously measured at single-cell resolution, thereby revealing the cellular and molecular basis of the pathogenesis of systemic lupus erythematosus. Combining specialized methods such as ACCENSE, FlowSOM, X-shift, Monocle 2, and PhenoGraph, CyTOF analysis has revealed the diversity of TLRs-induced specific responses within cell types, as well as the unique CD14 chromatin shared by clinically heterogeneous patients. hi Cytokine signatures of monocyte secretion. These studies illustrate the advantages of CyTOF for in-depth analysis of immune cell phenotypes and provide valuable insights into the mechanisms of systemic lupus erythematosus.
[0005] It is worth noting that the multidimensionality and complexity of CyTOF also includes a comprehensive understanding of the high-dimensional insights into the "DA network" between cell clusters. Insights into abnormalities in the interaction patterns between multiple cell clusters can advance the systematic identification and understanding of characteristics related to disease etiology, progression, and pathogenesis. Studies have shown that PD1 hi CXCR5 - Peripheral helper T cells can promote B cell responses and antibody production in autoimmune diseases. Understanding the dysregulation of interaction patterns among multiple cell clusters can improve the systematic identification and understanding of disease-associated features, rather than identifying a single dysregulated cell subset.
[0006] However, current high-dimensional data analysis methods cannot mine the hidden interaction patterns between cell clusters from such datasets. Although statistical correlation and significance between cell clusters can be easily achieved, the key challenge in mining high-dimensional single-cell proteomics data is to construct and quantify the global interaction network across cell clusters to understand the pathogenesis of multifactorial complex diseases such as SLE.
[0007] Therefore, technicians in this field are committed to developing a network analysis method for mass spectrometry data mining. The SCANCell analysis method converts the available information hidden in high-dimensional mass spectrometry data into a DA network between cell clusters; the DA network can intuitively reflect the normal or damaged function of the immune system; and the abnormal interaction patterns between healthy samples and disease samples can be mined from the network, which is conducive to the description of the pathogenesis of the disease. Summary of the invention
[0008] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is the SCANCell analysis method, which converts the available information hidden in high-dimensional mass spectrometry data into the topological structure of the DA network between cell clusters; provides a new dimension of understanding for the pathogenesis of the disease by considering the protein expression level of cells and the abundance of cell clusters; mines high-dimensional proteomic data to understand disease-specific cellular and molecular distortions.
[0009] To achieve the above object, the present invention provides a network analysis method SCANCell for mass spectrometry data mining, including the construction of a DA network for sample grouping and cell type.
[0010] Furthermore, the SCANCell divides cells of all samples into different cell clusters, and extracts corresponding cell abundance histograms in the clusters as features of each sample.
[0011] Furthermore, the SCANCell uses EMD to calculate the distance matrix between all samples, divides all samples into different subgroups, and constructs a DA network of all cell clusters for each subgroup.
[0012] Furthermore, the SCANCell utilizes partial mutual information to quantify the interaction strength between cell clusters.
[0013] Furthermore, in the DA network, each node represents a cell cluster, each edge represents the interaction between two cell clusters, and the weight of the edge represents the interaction strength.
[0014] Furthermore, the DA network was visualized by Cytoscape software.
[0015] Further, the following steps are included:
[0016] Step 1: Cell clustering;
[0017] Step 2: Sample grouping;
[0018] Step 3: Construction of DA network.
[0019] Furthermore, the step 1 comprises the following steps:
[0020] Step 1.1. All SLE data were normalized by an inverse hyperbolic sine (arcsinh) function;
[0021] Step 1.2: Use the completely unsupervised Monocle 2 algorithm to complete cell clustering, select a certain number of cells from each sample, and merge these selected cells into a whole cell data;
[0022] Step 1.3: Use the Monocle 2 algorithm to further divide the aggregated cells into several cell clusters, calculate the distance between all remaining cells and the center point of each cell cluster, and assign the remaining cells to the cell cluster with the closest distance.
[0023] Furthermore, the step 2 comprises the following steps:
[0024] Step 2.1: Extract effective features of each sample from the high-dimensional data structure and re-describe each sample as computable and actionable data;
[0025] Step 2.2: Use EMD to calculate the distance matrix between all samples, and then use this matrix as input to group the samples using the hierarchical clustering algorithm.
[0026] Furthermore, the step 3 comprises the following steps:
[0027] Step 3.1: quantify the direct association between two cell clusters and construct a global DA network of all cluster-cluster interactions;
[0028] Step 3.2, the DA network retains the direct interactions between cell clusters and weakens the indirect interactions;
[0029] Step 3.3: What describes the pathogenesis is the abnormal interaction pattern in the DA network, rather than simply the abnormal expression of a certain cell cluster.
[0030] In a preferred embodiment of the present invention, the present invention provides a data analysis model, in particular, a network analysis algorithm between cell subpopulations involving high-throughput mass spectrometry data.
[0031] The present invention has developed the SCANCell analysis method, which can convert the available information hidden in high-dimensional mass spectrometry data into the topological structure of the DA network between cell clusters. The DA network characterizes the strength of direct interactions between immune cell clusters and weakens the interference of indirect interactions, which enables SCANCell to quantitatively identify disease-specific abnormal interaction patterns at the DA network topological structure level. SCANCell provides a new perspective for understanding the pathogenesis of the disease by considering the abnormal interaction patterns of cell clusters. Using SCANCell to analyze the interactions between cell clusters in the entire disease spectrum of SLE, the present invention found that compared with patients with stable SLE, patients with active SLE revealed CD8 + The inter-cluster interactions of T cells are mitigated. The present invention provides a tool for deep mining of high-dimensional proteomic data to better understand disease-specific cellular and molecular aberrations. This is the first exploratory study to depict the pathogenesis of the disease from the perspective of inter-cluster interactions of cells, which can facilitate the identification of potential drug targets.
[0032] Taking the cell-protein molecule expression matrix data of all given samples as input, SCANCell uses the features extracted from the samples to divide all samples into different "subgroups" and constructs a DA network of all cell clusters for each subgroup. DA is calculated by a global nonlinear index to characterize the interaction strength of cell clusters in the entire immune system. Sample grouping is an important step in mining abnormal interaction patterns in the DA network, which helps to eliminate the interference of unnecessary cluster interactions on the DA network. In addition, sample grouping enables samples within subgroups to have their own outstanding representative characteristics, such as the enrichment of specific cell populations.
[0033] In general, each sample is formalized as a high-dimensional data structure consisting of heterogeneous cell clusters, each of which has dozens of protein expression features. Before grouping samples, a key step is to extract useful features of each sample from the currently available information. Based on this fact, SCANCell first divides the cells of all samples into different cell clusters, and extracts the corresponding cell abundance histograms in the clusters as the features of each sample. In principle, SCANCell uses the Euclidean distance between cell clusters and the cell cluster abundance of each sample, introduces EMD to measure the distance matrix between samples, and divides the samples of the same group into different subgroups. SCANCell represents each subgroup as a matrix of cell cluster-protein expression level and constructs its corresponding DA network. In the DA network, each node represents a cell cluster, and each edge represents the interaction between two cell clusters, where the weight of the edge represents the interaction strength. In order to enable the constructed network to directly reflect the interaction between immune cell clusters at the system level, SCANCell uses partial mutual information to quantify the interaction strength between cell clusters. Partial mutual information is a nonlinear measure calculated based on the probabilistic concept of partial independence, which not only considers the dependencies between pairs of cell clusters like cosine similarity and Euclidean distance, but also their association with the rest of the cell clusters. The constructed DA network was visualized by Cytoscape software.
[0034] The present invention proposes a network analysis method, which reveals that abnormal inter-cluster interactions may be related to the onset of systemic lupus erythematosus, and explains the therapeutic effect of interleukin 2 (IL-2) from the perspective of inter-cluster DA network.
[0035] The present study mined the inter-cell cluster DA network contained in the high-throughput data related to systemic lupus erythematosus, and provided an analytical method for immunobiologists to deeply explore the molecular-level disorder mechanism in the disease.
[0036] Compared with the prior art, the present invention has the following obvious substantial features and significant advantages:
[0037] 1. Developed a direct association network between sample groupings and cell clusters, SCANCell, a new analytical method that can convert available information hidden in high-dimensional mass spectrometry data into a DA network between cell clusters. SCANCell is the first method to construct a DA network of interactions between cell clusters and use it to identify disease-specific abnormal features, which is lacking in all existing analytical methods.
[0038] 2. Although no single cell cluster showed statistical differences among HCs, APs, and RPs, SCANCell analysis found that CD8 +The inter-cluster interactions of T cells were lower than those of HCs and RPs, which may be due to the fact that CD8 + The effector function of T cells is in an overactive state. In addition, it was found that CD8 + Intercluster interactions of T cells.
[0039] 3. Given that sample grouping maintains the homogeneous characteristics of samples within subgroups and the heterogeneous characteristics of samples between subgroups, and network construction weakens indirect interactions, the DA network can intuitively reflect the normal or impaired function of the immune system. Therefore, abnormal interaction patterns between healthy samples and disease samples can be mined from the network, which is conducive to revealing the pathogenesis of the disease.
[0040] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is the interactive differences of cell subpopulations in the whole sample and sample subgroups mined by the algorithm of a preferred embodiment of the present invention;
[0042] Figure 2 The algorithm of a preferred embodiment of the present invention mines the differences in the interactive network relationships of cell subpopulations in different patient groups. DETAILED DESCRIPTION
[0043] The following describes several preferred embodiments of the present invention with reference to the drawings in the specification, so that the technical content is clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned in the text.
[0044] In the drawings, components with the same structure are indicated by the same numerical reference numerals, and components with similar structures or functions are indicated by similar numerical reference numerals. The size and thickness of each component shown in the drawings are arbitrarily shown, and the present invention does not limit the size and thickness of each component. In order to make the illustration clearer, the thickness of the components is appropriately exaggerated in some places in the drawings.
[0045] A novel network analysis model for mass spectrometry streaming data mining.
[0046] Step 1: Cell clustering.
[0047] 1. Before cell clustering, all SLE data were normalized by an inverse hyperbolic sine (arcsinh) function.
[0048] 2. Here, a completely unsupervised Monocle 2 algorithm is used to complete the task of cell clustering. It first selects a certain number of cells from each sample and merges these selected cells.
[0049] 3. The aggregated cells are further divided into several cell clusters by applying the Monocle 2 algorithm. Then, the remaining cells are assigned to the cell cluster with the closest distance by calculating the distance between all remaining cells and the center point of each cell cluster. In other words, in SCANCell, we analyze all cells of the SLE sample, not just some of the cells after sampling.
[0050] Step 2: Sample grouping.
[0051] 1. Extract effective features of each sample from the high-dimensional data structure and re-describe each sample as computable and actionable data.
[0052] 2. Use EMD to calculate the distance matrix between all samples, and then use this matrix as input to group the samples using the hierarchical clustering algorithm.
[0053] Step 3: Construction of direct interaction network.
[0054] 1. Quantify the direct association between two clusters and construct a DA network of all cluster-cluster interactions.
[0055] 2. This network retains direct interactions between cell clusters as much as possible while weakening indirect interactions, playing an important role in the analysis of disease pathogenesis.
[0056] 3. The normal function of the entire immune system is maintained by the synergistic action of various cell clusters, so the abnormal interaction patterns in the network can more accurately reflect the disease-related pathogenesis rather than simply the abnormal expression of a certain cell cluster.
[0057] like Figure 1 As shown, the DA network of CD3+T cells in patients with active systemic lupus erythematosus. Figure (left): Construction of DA network of cell clusters for all active lupus patients (APs) without sample grouping, Figure (middle): Construction of DA network of cell clusters for active lupus patients in subgroup 1 (AP1), Figure (right): Construction of DA network of cell clusters for active lupus patients in subgroup 2 (AP2).
[0058] like Figure 2As shown, the DA network constructed by SCANCell for each subgroup. Subgroup 1 of the normal group (HC1, upper left), subgroup 2 of the normal group (HC2, lower left); subgroup 1 of the active lupus patient group (AP1, upper middle), subgroup 2 of the active lupus patient group (AP2, lower middle); subgroup 1 of the stable lupus patient group (RP1, upper right), subgroup 2 of the stable lupus patient group (RP2, lower right). In the DA network, numbered circles represent cell clusters, and the thickness of the edges indicates the strength of the interaction between clusters. Light-colored cell clusters indicate different interaction patterns in the two subgroups. For example, Cluster 2 in HC1 is included in the complete graph constructed by Cluster 2, Cluster 8, and Cluster 9, while Cluster 2 in HC2 is different.
[0059] The preferred specific embodiments of the present invention are described in detail above. It should be understood that ordinary technicians in the field can make many modifications and changes based on the concept of the present invention without creative changes. Therefore, all technical solutions that can be obtained by technicians in the technical field based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the scope of protection determined by the claims.
Claims
1. A network analysis method for mass spectrometry data mining, characterized in that: Sample grouping and construction of DA network between cell clusters, wherein the DA network is a directly associated network; All samples were divided into different subgroups using the features extracted from them, and the DA network of all cell clusters was constructed for each subgroup; The cells of all samples are divided into different cell clusters, and the corresponding cell abundance histograms in the clusters are extracted as the features of each sample; The strength of interactions between cell clusters was quantified using partial mutual information; In the DA network, each node represents a cell cluster, each edge represents the interaction between two cell clusters, and the weight of the edge represents the interaction strength; The DA network was visualized by Cytoscape software.
2. The network analysis method for mass spectrometry streaming data mining according to claim 1, characterized in that: The steps include: Step 1: Cell clustering; Step 2: Sample grouping; Step 3: Construction of DA network.
3. The network analysis method for mass spectrometry streaming data mining according to claim 2, characterized in that: The step 1 comprises the following steps: Step 1.
1. All SLE data were normalized by an inverse hyperbolic sine function; Step 1.2: Use the completely unsupervised Monocle 2 algorithm to complete cell clustering, select a certain number of cells from each sample, and merge these selected cells into a whole cell data; Step 1.3: Apply the Monocle 2 algorithm to further divide the aggregated cells into different cell clusters, calculate the distance between all remaining cells and the center point of each cell cluster, and assign the remaining cells to the cell cluster with the closest distance.
4. The network analysis method for mass spectrometry streaming data mining according to claim 2, characterized in that: The step 2 comprises the following steps: Step 2.1: Extract effective features of each sample from the high-dimensional data structure and re-describe each sample as computable and actionable data; Step 2.2: Use the earth mover distance to calculate the distance matrix between all samples, and then use this matrix as input to group the samples using the hierarchical clustering algorithm.
5. The network analysis method for mass spectrometry streaming data mining according to claim 2, characterized in that: The step 3 comprises the following steps: Step 3.1: quantify the direct association between two cell clusters and construct a global DA network of all cluster-cluster interactions; Step 3.2, the DA network reflects the direct interaction between cell clusters and weakens the indirect interaction; Step 3.3: Describe the pathogenesis based on abnormal interaction patterns in the network, rather than simply abnormal expression of a certain cell cluster.