A biological relationship network information modeling method and system

Abstract

The application discloses a biological relationship network information modeling method and system. The method comprises the following steps: constructing a graph structure biological network by using omics data, wherein the nodes of the biological network reflect the omics data categories, the edges reflect the correlation between the nodes, and the nodes of the biological network comprise expression profile nodes and non-expression profile nodes; calculating a relative relationship matrix based on the biological network, wherein the relative relationship matrix reflects the relative values with normal samples; performing dimension reduction on the relative relationship matrix by using a self-encoder to obtain reduced features; screening the reduced features according to the importance of feature expression, and then performing diagnosis and prognosis analysis by using the screened features. The application can effectively improve the accuracy and application range of biological omics data analysis.

Description

Technical Field

[0001] This invention relates to the field of medical component analysis technology, and more specifically, to a method and system for modeling biological relationship network information. Background Technology

[0002] Research indicates that human diseases are not caused by defects in a single molecule, but rather by complex interactions between multiple molecular mediators. The use of graphs and / or biological networks to analyze human diseases plays a crucial role in precision oncology. Currently, researchers commonly use graphs to represent complex networks and employ graph theory methods to study them. In biomolecular networks, nodes represent biomolecules, and edges represent various relationships such as regulation and interactions between biomolecules. Biomolecules include genes and proteins.

[0003] Biological networks can be broadly categorized into two types: evidence-based analytical networks and statistical inference networks. The first type relies on experimental evidence of specific analytical interactions, such as PPI networks, metabolic networks, and regulatory networks (transcription factor-gene networks, non-coding RNA-gene networks). The second type is statistical inference networks, which are based on statistical inference and depend on interactions between components established through statistical analysis. Examples include "co-expression networks," where genes are connected based on statistically significant or inversely correlated (depending on the underlying problem) expression profiles related to a disease of interest, such as WGCNA. Another type of statistically generated network is the "genetic network."

[0004] Network analysis primarily focuses on identifying network modules or key nodes. Current methods also exist for identifying cancer-driving modules and biomarkers based on biological networks, such as methods for identifying cancer-driving modules based on biological networks and subcellular localization data (e.g., patent application publication number CN112259163). Researchers typically select important modules or network signatures, construct subnetworks based on the network's nodes and edges, and establish connections between these subnetworks and disease characteristics to obtain important "modules" representing specific diseases. Then, subsequent problem analysis is performed based on the expression information of nodes within these modules, or the network is transformed into a matrix based on the mutual information between nodes for subsequent cancer feature analysis, such as prognostic subtype clustering.

[0005] While biological network analysis is gaining increasing attention and is frequently used to analyze disease-related issues in current technologies, these analyses typically suffer from the following problems:

[0006] (1) Ignoring network information. Most methods for analyzing biological networks only utilize nodes that are of significant importance in the network or nodes in modules, while ignoring the interactions between nodes in the network. Therefore, these methods can only be called node set-based methods, rather than true network analysis methods.

[0007] (2) Complete node expression profiles are required. For network analysis, each node must have an expression profile; nodes without a profile will be ignored in the network. Therefore, due to certain issues and data completeness requirements, or when analyzing publicly available data, large-volume or comprehensive omics datasets are often used. This leads to the neglect of some difficult-to-prepare omics datasets or those less frequently included in public databases. For example, obtaining metabolomics data requires expensive instruments and involves variable sensitivity and high volatility; it is also rarely collected in large-scale multi-omics projects, such as The Tumor Genome Atlas (TCGA).

[0008] (3) Lack of consideration for redundant information. In the analysis of node relationships, the analysis of node relationships that would increase dimensionality, increase the amount of information and increase the computational burden, or make the information redundant are ignored, because in a network, one node usually interacts with multiple nodes. However, this point is ignored in the research.

[0009] The aforementioned technical issues limit the methods available for analyzing actual clinical situations. According to currently published literature, there is still no solution to construct a relational matrix based on missing expression profiles of some nodes in the network and to reduce redundant information. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for modeling biological relationship network information.

[0011] According to a first aspect of the present invention, a method for modeling biological relationship networks is provided. The method includes the following steps:

[0012] A graph-structured biological network is constructed using omics data. The nodes of this biological network reflect the omics data categories, the edges reflect the correlations between the nodes, and the nodes of the biological network include expression profile nodes and non-expression profile nodes.

[0013] A relative relationship matrix is ​​calculated based on the biological network, and the relative relationship matrix reflects the relative value with normal samples;

[0014] The relative relationship matrix is ​​reduced in dimensionality using an autoencoder to obtain the reduced features.

[0015] Based on the importance of the feature representation, the dimensionality-reduced features are screened, and then the screened features are used for diagnosis and prognostic analysis.

[0016] According to a second aspect of the present invention, a biological relationship network information modeling system is provided. The system includes:

[0017] Biological Network Construction Module: Used to construct a graph-structured biological network using omics data. The nodes of the biological network reflect the omics data categories, the edges reflect the correlations between the nodes, and the nodes of the biological network include expression profile nodes and non-expression profile nodes.

[0018] Relationship matrix representation module: used to calculate a relative relationship matrix based on the biological network, the relative relationship matrix reflecting the relative value with normal samples;

[0019] Feature dimensionality reduction module: used to reduce the dimensionality of the relative relationship matrix using an autoencoder to obtain the dimensionality-reduced features;

[0020] Diagnosis and prognosis analysis module: This module is used to filter the dimensionality-reduced features based on their importance, and then use the filtered features for diagnosis and prognosis analysis.

[0021] Compared with the prior art, the advantages of the present invention are that the biological relationship network information modeling method provided represents the relevance of all nodes according to the connections of nodes in the network, and combines the node expression spectrum to convert it into a node relevance matrix. Then, machine learning methods are used to reduce the dimensionality of the matrix features to reduce redundant features. The dimensionality-reduced features are then used for subsequent clinical problem analysis, thereby solving the problems of utilizing biological network information and removing redundant information.

[0022] Other features and advantages of the invention will become clear from the following detailed description of exemplary embodiments of the invention with reference to the accompanying drawings. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments of the invention and, together with their description, serve to explain the principles of the invention.

[0024] Figure 1 This is a flowchart of a biological relationship network information modeling method according to an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram illustrating the process of modeling biological relationship network information according to an embodiment of the present invention. Detailed Implementation

[0026] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0027] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0028] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0029] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0030] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0031] Combination Figure 1 and Figure 2 As shown, the provided method for modeling biological relationship networks includes the following steps:

[0032] Step S110: Preprocess the biological omics data and determine the corresponding expression profile information.

[0033] Based on the problem to be analyzed, select omics data and determine whether the omics data has an expression profile. If so, perform normalization processing. In this way, the categories of biological omics data can be used as network nodes, and the expression profile information of each node can be confirmed, including both cases with and without expression profiles.

[0034] Step S120: Construct a graph-structured biological network and determine the node types and the connections between nodes.

[0035] Specifically, the connections between nodes are determined based on the biological relationship nodes in an open access database to construct an omics network, also known as a biological network. This network is undirected. For example, based on the correlation between proteins, a protein-protein network (PPI) can be constructed; based on the reaction relationship between proteins and metabolites, a metabolic protein network (MPI) can be constructed; and based on the regulatory relationship between mRNA and non-coding RNA, a regulatory network can be constructed.

[0036] Therefore, the constructed biological network nodes can be mRNA, proteins, metabolites, or other types. Furthermore, edge connections are established based on the correlations between nodes, including biological correlations, expression correlations, structural correlations, or other types of correlations.

[0037] Preferably, the similarity between nodes can also be incorporated to adjust the network of the node connections, such as based on the similarity between node structures or expressions.

[0038] Step S130: Calculate the relative relationship matrix reflecting the relationships between nodes based on the constructed biological network.

[0039] In this step S130, a relationship matrix is ​​constructed based on the connectivity structure of the biological network and the expression profile information, specifically including the following steps:

[0040] Step S131: Calculate the network relation vector

[0041] For example, the correlation score RS between connected nodes a and b in a network is calculated to characterize their biological connection, as follows:

[0042] RS = Log2(m+2)

[0043] Where m represents the intermediate node c i The number of nodes, i.e., nodes a and b can be connected via a common path c. i Direct connections. In addition, a type of omics information interacts not only with information within the same omics but also with other types of omics information. For example, proteins interact with each other, and proteins also interact with metabolites. Therefore, m is not limited to one type of omics node.

[0044] Step S132, calculate the relation matrix

[0045] For example, by multiplying the correlation of edges in the network (i.e., the correlation of connections between nodes) with the expression values ​​of the corresponding nodes with expression spectra, the node expression spectra are transformed into an edge matrix, thereby increasing the number of features. That is, the columns of the matrix are samples, the rows of the matrix are transformed from nodes to edges connecting nodes, and the elements in the matrix represent the quantitative relationship of the edges in the network, multiplied by the corresponding node expression spectra.

[0046] Step S133: Calculate the relative relationship matrix.

[0047] Since the expression of omics information typically differs between normal and tumor tissues, a relationship matrix between tumor and normal tissues is calculated. First, the mean of the normal samples in the edge relationship matrix is ​​calculated, resulting in a mean vector of the edges formed by the connected nodes of the normal samples. Then, the elements corresponding to the same edges in the vectors for each cancer sample in the edge relationship matrix are subtracted. Finally, the row mean of the identical edge features is taken, thus obtaining a relative relationship matrix of network nodes relative to normal tissues (i.e., a relative relationship matrix).

[0048] It should be noted that the existence of identical repeated edges is due to the consideration of missing expression spectra for some nodes. Therefore, after the network is built, the following connection situations will exist: nodes without expression spectra are connected to nodes with expression spectra; nodes with expression spectra are connected to nodes with expression spectra; nodes without expression spectra are connected to nodes without expression spectra.

[0049] Because the relative relationship matrix calculated in this invention also requires the participation of node expression spectra, when nodes without expression spectra are connected, they are considered as intermediate nodes when calculating the correlation between nodes, but are ignored when calculating the correlation matrix (i.e., the relationship matrix and the relative relationship matrix).

[0050] When nodes with expression spectra are connected to each other, the correlation element of the same edge will be multiplied by the expression spectra of the two nodes respectively. Therefore, when calculating the correlation matrix in this case, the row mean of the same edge is taken.

[0051] In summary, in step S130, the conversion of the network node connection relationship and the node expression spectrum into an edge relationship matrix is ​​beneficial for better characterizing the correlation between nodes.

[0052] Step S140: Use an autoencoder to perform feature dimensionality reduction on the relative relation matrix to obtain the dimensionality-reduced features.

[0053] An autoencoder is an unsupervised, feedforward, non-recurrent neural network that consists of an input layer, multiple hidden layers, and an output layer. For example, the input layer takes x of dimension n as input, transforms x through successive hidden layers, and outputs a reconstructed x' of x, where x' has the same dimension as x.

[0054] In one embodiment, an autoencoder with three hidden layers (512, 256, and 516 nodes respectively) was implemented using the PyTorch framework, with a learning rate of 0.001. The relational matrix of cancer samples was used as input to the autoencoder framework, and the bottleneck layer of the autoencoder generated new features from the omics data. The batch size and epochs were set for model training. An epoch refers to the iteration of the learning algorithm over the entire training dataset, and the batch size is the number of samples trained in each iteration.

[0055] Step S150 involves further feature screening of the dimensionality-reduced features, followed by diagnosis and prognostic analysis.

[0056] After reducing the initial number of features to a few new features obtained from the bottleneck layer, the autoencoder can perform feature selection based on the specific problem being analyzed. For example, it can select prognostic-related features through univariate Cox analysis to further ensure the effectiveness of the features for the analytical problem. The selected new features can then be used for prognostic analysis of disease problems or classification analysis and prediction of disease features. Feature selection can retain important features and improve computational efficiency with almost no impact on the representation effect.

[0057] Accordingly, the present invention also provides a biological relationship network information modeling system for implementing one or more aspects of the above-described method. For example, the system includes: a data preprocessing module for preprocessing biological omics data and determining corresponding expression profile information; a biological network construction module for constructing a graph-structured biological network using the omics data, wherein the nodes of the biological network reflect the omics data categories, the edges reflect the correlations between nodes, and the nodes of the biological network include expression profile nodes and non-expression profile nodes; a relationship matrix representation module for calculating a relative relationship matrix based on the biological network, wherein the relative relationship matrix reflects the relative value with normal samples; a feature dimensionality reduction module for reducing the dimensionality of the relative relationship matrix using an autoencoder to obtain dimensionality-reduced features; and a diagnosis and prognostic analysis module for screening the dimensionality-reduced features according to the importance of feature expression, and then using the screened features for diagnosis and prognostic analysis. Each module involved can be implemented using software, dedicated hardware, a processor, or an FPGA.

[0058] Without departing from the spirit and scope of this invention, those skilled in the art can make appropriate modifications to the above embodiments. For example, other autoencoder structures can be used. This invention does not limit the specific structure, number of layers, training parameters, etc. of the autoencoder.

[0059] In summary, the present invention has the following technical effects:

[0060] 1) This invention combines biological network information with expression profiles and utilizes deep learning methods to reduce redundant information, enabling its application in analyzing cancer characteristics and clinical problems. On one hand, by transforming the relationship matrix, the original features are expanded, enriching the original network node expression profile features. On the other hand, deep learning is used for feature dimensionality reduction, reducing redundant features. Therefore, it balances the effectiveness of feature representation with computational efficiency.

[0061] 2) This invention addresses the issue of missing expression profiles for certain nodes in biological networks. The lack of expression profiles leads to the neglect of some omics or nodes in research. However, in this invention, for some nodes, even without expression profiles, node information can still be characterized based on the network for feature analysis. For example, metabolomics plays a crucial role in cancer development; metabolite-protein networks can be constructed to analyze cancer characteristics and explore the specific impact of metabolism on metabolites and cancer development.

[0062] 3) This invention analyzes cancer features by fusing network node relation representations with node expression spectra to form a network edge relation matrix. This not only integrates information features from multiple omics nodes but also avoids the limitation that all node expression spectra must be complete, and it can expand upon the original features. Furthermore, it utilizes computer and information technology to reduce redundant features and retain important ones, enriching the information available for cancer analysis and exploring the impact of multi-level information on cancer development.

[0063] This invention can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of the invention.

[0064] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0065] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0066] The computer program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, Python, etc., and conventional procedural programming languages ​​such as "C" or similar languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of the invention.

[0067] Various aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0068] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0069] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0070] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions. It will be known to those skilled in the art that implementation in hardware, implementation in software, and implementation using a combination of software and hardware are equivalent.

[0071] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, and are not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the invention is defined by the appended claims.

Claims

1. A method for modeling biological relationship networks, comprising the following steps: A graph-structured biological network is constructed using omics data. The nodes of this biological network reflect the omics data categories, the edges reflect the correlations between the nodes, and the nodes of the biological network include expression profile nodes and non-expression profile nodes. A relative relationship matrix is ​​calculated based on the biological network, which reflects the correlation between nodes and the relative value of normal samples; The relative relationship matrix is ​​reduced in dimensionality using an autoencoder to obtain the reduced features. Based on the importance of the feature representation, the dimensionality-reduced features are screened, and then the screened features are used for diagnosis and prognostic analysis. The relative relationship matrix is ​​constructed according to the following steps: The correlation score RS between nodes a and b in the biological network is calculated to characterize the biological connection between nodes a and b, and is expressed as follows: In this case, node a and node b are connected through a common node. Direct connection, where m represents an intermediate node in the biological network. The number of; The correlation of edges in the biological network is multiplied by the expression values ​​of the corresponding nodes with expression profiles to transform the node expression profiles into an edge relation matrix. Calculate the relative relationship matrix of the edge relationship matrix with respect to the normal organization; The biological network includes nodes with no expression profile connected to nodes with expression profile, nodes with expression profile connected to nodes with expression profile, and nodes with no expression profile connected to nodes with no expression profile, wherein: When nodes with no expression profiles are connected to each other, they are considered as intermediate nodes when calculating the correlation between nodes, and are ignored when calculating the relationship matrix. For nodes with expression spectra connected to each other, the correlation element of the same edge is multiplied by the expression spectra of the two nodes respectively, and the average value of the same edge in the relation matrix is ​​calculated.

2. The method according to claim 1, characterized in that, Calculating the relative relationship matrix of the edge relationship matrix with respect to the normal organization includes: The mean of the normal samples in the edge relation matrix is ​​taken to obtain the relation mean vector of the edges formed by the connected nodes of the normal samples. Subtract the elements of the vector that correspond to the same edge from each cancer sample in the edge relation matrix; By taking the row mean of the same repeated edge features, a relative relationship matrix of network nodes relative to a normal organization is obtained.

3. The method according to claim 1, characterized in that, The omics data categories include mRNA, protein, or metabolites, and the correlations between the nodes include biological correlations, expression correlations, or structural similarities.

4. The method according to claim 1, characterized in that, The omics data were preprocessed through the following steps: Select omics data based on the questions to be analyzed; Determine whether the omics data has an expression profile. If so, perform normalization processing, use the categories of biological omics as network nodes, and obtain the expression profile information of each node.

5. A biological relationship network information modeling system, comprising: Biological Network Construction Module: Used to construct a graph-structured biological network using omics data. The nodes of the biological network reflect the omics data categories, the edge connections reflect the correlations between the nodes, and the nodes of the biological network include expression profile nodes and non-expression profile nodes. Relationship matrix representation module: used to calculate a relative relationship matrix based on the biological network, the relative relationship matrix reflecting the relative value with normal samples; Feature dimensionality reduction module: used to reduce the dimensionality of the relative relationship matrix using an autoencoder to obtain the dimensionality-reduced features; Diagnosis and prognosis analysis module: This module is used to filter the dimensionality-reduced features based on the importance of the feature representation, and then use the filtered features for diagnosis and prognosis analysis. The relative relationship matrix is ​​constructed according to the following steps: The correlation score RS between nodes a and b in the biological network is calculated to characterize the biological connection between nodes a and b, and is expressed as follows: In this case, node a and node b are connected through a common node. Direct connection, where m represents an intermediate node in the biological network. The number of; The correlation of edges in the biological network is multiplied by the expression values ​​of the corresponding nodes with expression profiles to transform the node expression profiles into an edge relation matrix. Calculate the relative relationship matrix of the edge relationship matrix with respect to the normal organization; The biological network includes nodes with no expression profile connected to nodes with expression profile, nodes with expression profile connected to nodes with expression profile, and nodes with no expression profile connected to nodes with no expression profile, wherein: When nodes with no expression profiles are connected to each other, they are considered as intermediate nodes when calculating the correlation between nodes, and are ignored when calculating the relationship matrix. For nodes with expression spectra connected to each other, the correlation element of the same edge is multiplied by the expression spectra of the two nodes respectively, and the average value of the same edge in the relation matrix is ​​calculated.

6. A computer-readable storage medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images